Research_Paper_Simplification.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zSXPSRi2gP_u"
      },
      "source": [
        "Loading Train Data"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "WVi_Z83fgP_u",
        "outputId": "b35516ac-351e-4336-a416-cc9a340e9c67"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Arrays loaded from Datasetfinalhope.txt\n"
          ]
        }
      ],
      "source": [
        "# Specify the file path from where to load the arrays\n",
        "file_path = \"Datasetfinalhope.txt\"\n",
        "\n",
        "# Initialize empty arrays to store loaded data\n",
        "input_texts = []\n",
        "target_texts = []\n",
        "\n",
        "# Open the file in read mode and load the arrays from the file\n",
        "with open(file_path, \"r\") as file:\n",
        "    for line in file:\n",
        "        # Split each line into input and target using the tab (\"\\t\") as a separator\n",
        "        input_text, target_text = line.strip().split(\"\\t\")\n",
        "\n",
        "        # Append the data to the respective arrays\n",
        "        input_texts.append(input_text)\n",
        "        target_texts.append(target_text)\n",
        "\n",
        "# Print a message indicating that arrays are loaded\n",
        "print(f\"Arrays loaded from {file_path}\")"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "Input Texts contain original sentences and Target Texts contain simplified sentences"
      ],
      "metadata": {
        "id": "TGwqgj19h_lU"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "zzzC54cDgP_w",
        "outputId": "9b703465-4f3a-4a82-ba1e-621a4a191665"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "4009"
            ]
          },
          "execution_count": 2,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "len(input_texts)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "SN3xM8qtgP_w",
        "outputId": "760d4547-cd5e-4ef1-b885-5956e82e4957"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "'The Wakker Prize (in German: Wakkerpreis, in French Prix Wakker, in Italian Premio Wakker) is awarded annually by the Swiss Heritage Society to a Municipality of Switzerland for the development and preservation of its architectural heritage.'"
            ]
          },
          "execution_count": 3,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "input_texts[3555]"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "t7HNIODOgP_w",
        "outputId": "15b11952-d236-42b3-c9cd-1cf5b2452136"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "'The Wakker Prize is awarded every year by the Swiss Heritage Society to a Municipality of Switzerland for the development and preservation of its architectural heritage.'"
            ]
          },
          "execution_count": 4,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "target_texts[3555]"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gQnMZ6uzgP_x"
      },
      "source": [
        "Preprocessing"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Jr5c7tyXgP_x",
        "outputId": "1621d6e0-471f-41aa-be2b-dd6b21c552a5"
      },
      "outputs": [
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "2023-12-15 03:32:43.846021: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  SSE4.1 SSE4.2 AVX AVX2 FMA\n",
            "To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\n"
          ]
        }
      ],
      "source": [
        "import tensorflow_datasets as tfds\n",
        "import tensorflow as tf\n",
        "from tensorflow.keras.preprocessing.text import Tokenizer\n",
        "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
        "import numpy as np\n",
        "import pickle\n",
        "\n",
        "# Tokenize the text data\n",
        "tokenizer = Tokenizer(filters='!\"#$%&()*+,-./:;<=>?@[\\\\]^_`{|}~\\t\\n', oov_token=\"<OOV>\", lower=True)\n",
        "tokenizer.fit_on_texts(input_texts + target_texts)\n",
        "\n",
        "# Save the tokenizer for later use during testing\n",
        "with open('tokenizer.pickle', 'wb') as handle:\n",
        "    pickle.dump(tokenizer, handle, protocol=pickle.HIGHEST_PROTOCOL)\n",
        "\n",
        "# Add the padding token to the word_index\n",
        "padding_token_index = len(tokenizer.word_index) + 1\n",
        "tokenizer.word_index['<PAD>'] = padding_token_index\n",
        "tokenizer.index_word[padding_token_index] = '<PAD>'\n",
        "\n",
        "# Convert text to sequences\n",
        "input_sequences = tokenizer.texts_to_sequences(input_texts)\n",
        "target_sequences = tokenizer.texts_to_sequences(target_texts)\n",
        "\n",
        "# Padding sequences\n",
        "max_sequence_length = 140  # Define the maximum sequence length\n",
        "encoder_input_data = pad_sequences(input_sequences, maxlen=max_sequence_length, padding='post')\n",
        "decoder_input_data = pad_sequences(target_sequences, maxlen=max_sequence_length, padding='post')\n",
        "\n",
        "# Generate decoder target data\n",
        "decoder_target_data = np.zeros_like(decoder_input_data)\n",
        "decoder_target_data[:, 0:-1] = decoder_input_data[:, 1:]\n",
        "decoder_target_data[:, -1] = padding_token_index  # Use the index of the padding token\n",
        "\n",
        "# Check Vocabulary Size Matching\n",
        "input_vocab_size = len(tokenizer.word_index) + 1\n",
        "target_vocab_size = len(tokenizer.word_index) + 1\n",
        "\n",
        "if input_vocab_size != target_vocab_size:\n",
        "    raise ValueError(\"Input and target vocabulary sizes do not match!\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "tjSyrB_dgP_x",
        "outputId": "1ed2f26e-3e09-4dc2-a60b-ef4f4acf2fa0"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "10112"
            ]
          },
          "execution_count": 6,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "input_vocab_size"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "-2DhvA9wgP_y",
        "outputId": "e949ec4a-1645-444c-fc0b-dd4ddcf66af9"
      },
      "outputs": [
        {
          "data": {
            "text/plain": [
              "10112"
            ]
          },
          "execution_count": 7,
          "metadata": {},
          "output_type": "execute_result"
        }
      ],
      "source": [
        "target_vocab_size"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RapIbuGxgP_y"
      },
      "source": [
        "Using Pretrained Word2Vec"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "nPm0qIzDgP_y",
        "outputId": "30c8198a-15a0-4de7-e84e-2f258bc64e5d"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Word2Vec Coverage: 79.14%\n"
          ]
        }
      ],
      "source": [
        "from gensim.models import KeyedVectors\n",
        "from keras.layers import Embedding\n",
        "import numpy as np\n",
        "\n",
        "# Specify the path to the Word2Vec model file\n",
        "word2vec_model_path = 'GoogleNews-vectors-negative300.bin'\n",
        "\n",
        "# Load the Google News Word2Vec model\n",
        "word2vec_model = KeyedVectors.load_word2vec_format(word2vec_model_path, binary=True)\n",
        "\n",
        "# Check Word2Vec Coverage\n",
        "covered_words = sum(1 for word in tokenizer.word_index if word in word2vec_model)\n",
        "total_words = len(tokenizer.word_index)\n",
        "\n",
        "coverage_percentage = (covered_words / total_words) * 100\n",
        "\n",
        "print(\"Word2Vec Coverage: {:.2f}%\".format(coverage_percentage))\n",
        "\n",
        "# Get the number of unique tokens from your tokenizer\n",
        "num_encoder_tokens = len(tokenizer.word_index) + 1  # Add 1 for the zero-padding token\n",
        "\n",
        "# Create an embedding matrix\n",
        "embedding_matrix = np.zeros((num_encoder_tokens, word2vec_model.vector_size))\n",
        "for word, i in tokenizer.word_index.items():\n",
        "    if word in word2vec_model:\n",
        "        embedding_matrix[i] = word2vec_model[word]\n",
        "\n",
        "# Create the embedding layer\n",
        "encoder_embedding_layer = Embedding(\n",
        "    num_encoder_tokens,\n",
        "    word2vec_model.vector_size,\n",
        "    weights=[embedding_matrix],\n",
        "    input_length=max_sequence_length,\n",
        "    trainable=False\n",
        ")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "pr4G7qNGgP_z"
      },
      "outputs": [],
      "source": [
        "import tensorflow as tf\n",
        "from tensorflow.keras.layers import Input, LSTM, Dense, Embedding, Bidirectional, Concatenate, Attention, GlobalMaxPooling1D, LayerNormalization, Dropout, Add\n",
        "from tensorflow.keras.models import Model\n",
        "from tensorflow.keras.callbacks import EarlyStopping\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "nsldG2FVgP_z",
        "outputId": "d45129b9-5986-4901-f2cf-085b28bc49bf"
      },
      "outputs": [
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "2023-12-15 03:33:35.567703: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  SSE4.1 SSE4.2 AVX AVX2 FMA\n",
            "To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\n",
            "2023-12-15 03:33:35.687843: I tensorflow/core/common_runtime/process_util.cc:146] Creating new thread pool with default inter op setting: 2. Tune using inter_op_parallelism_threads for best performance.\n"
          ]
        }
      ],
      "source": [
        "# Convert data to tensors\n",
        "encoder_input_data = tf.convert_to_tensor(encoder_input_data, dtype=tf.float32)\n",
        "decoder_input_data = tf.convert_to_tensor(decoder_input_data, dtype=tf.float32)\n",
        "decoder_target_data = tf.convert_to_tensor(decoder_target_data, dtype=tf.float32)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "N_sxj2YQgP_z",
        "outputId": "820d4406-df9f-4135-c74e-d1c6a66d7be6"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Encoder Input Data Shape: (4009, 140)\n",
            "Decoder Input Data Shape: (4009, 140)\n",
            "Decoder Target Data Shape: (4009, 140)\n"
          ]
        }
      ],
      "source": [
        "print(\"Encoder Input Data Shape:\", encoder_input_data.shape)\n",
        "print(\"Decoder Input Data Shape:\", decoder_input_data.shape)\n",
        "print(\"Decoder Target Data Shape:\", decoder_target_data.shape)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "p8u6yPcDgP_z"
      },
      "outputs": [],
      "source": [
        "# Define the parameters\n",
        "latent_dim = 120\n",
        "num_encoder_tokens = len(tokenizer.word_index) + 1  # Adjust according to your data\n",
        "num_decoder_tokens = len(tokenizer.word_index) + 1  # Adjust according to your data"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EIItPl29gP_z"
      },
      "source": [
        "Training the Model (1)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "_n0UoOJSgP_0"
      },
      "outputs": [],
      "source": [
        "from tensorflow.keras.layers import Input, Embedding, Bidirectional, LSTM, Concatenate, Attention, Dense, LayerNormalization, Dropout, Add\n",
        "from tensorflow.keras.models import Model\n",
        "\n",
        "# Define the encoder\n",
        "encoder_inputs = Input(shape=(None,))\n",
        "encoder_embedding = encoder_embedding_layer(encoder_inputs)  # Assuming encoder_embedding_layer is already defined\n",
        "encoder_bi_lstm = Bidirectional(LSTM(latent_dim, return_sequences=True, return_state=True))\n",
        "encoder_outputs, forward_h, forward_c, backward_h, backward_c = encoder_bi_lstm(encoder_embedding)\n",
        "state_h = Concatenate()([forward_h, backward_h])\n",
        "state_c = Concatenate()([forward_c, backward_c])\n",
        "\n",
        "# Define the decoder\n",
        "decoder_inputs = Input(shape=(None,))\n",
        "decoder_embedding = Embedding(num_decoder_tokens, latent_dim, mask_zero=True)(decoder_inputs)\n",
        "\n",
        "# Update the initial state to match the expected state_size of the LSTM cell\n",
        "decoder_lstm = LSTM(latent_dim*2, return_sequences=True, return_state=True)\n",
        "decoder_outputs, _, _ = decoder_lstm(decoder_embedding, initial_state=[state_h, state_c])\n",
        "\n",
        "attention = Attention()([decoder_outputs, encoder_outputs])\n",
        "decoder_concat = Concatenate(axis=-1)([decoder_outputs, attention])\n",
        "decoder_dense_2 = Dense(latent_dim*2, activation='relu')\n",
        "decoder_norm = LayerNormalization()\n",
        "decoder_dropout = Dropout(0.5)\n",
        "decoder_outputs_2 = decoder_dense_2(decoder_concat)\n",
        "decoder_outputs_2_norm = decoder_norm(decoder_outputs_2)\n",
        "decoder_outputs_2_dropout = decoder_dropout(decoder_outputs_2_norm)\n",
        "\n",
        "# Ensure the dimensions are compatible during the residual connection\n",
        "decoder_residual = Add()([decoder_outputs, decoder_outputs_2_dropout])\n",
        "\n",
        "# Define the final Dense layer\n",
        "final_decoder_dense = Dense(num_decoder_tokens, activation='softmax')\n",
        "decoder_outputs_final = final_decoder_dense(decoder_residual)\n",
        "\n",
        "# Define the model\n",
        "model = Model([encoder_inputs, decoder_inputs], decoder_outputs_final)\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "wdubr5ZTgP_0",
        "outputId": "500edb06-c74c-427b-fec5-ca1f6d9a3876"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Model: \"model_1\"\n",
            "__________________________________________________________________________________________________\n",
            " Layer (type)                   Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            " input_3 (InputLayer)           [(None, None)]       0           []                               \n",
            "                                                                                                  \n",
            " embedding (Embedding)          (None, None, 300)    3033600     ['input_3[0][0]']                \n",
            "                                                                                                  \n",
            " input_4 (InputLayer)           [(None, None)]       0           []                               \n",
            "                                                                                                  \n",
            " bidirectional_1 (Bidirectional  [(None, None, 240),  404160     ['embedding[1][0]']              \n",
            " )                               (None, 120),                                                     \n",
            "                                 (None, 120),                                                     \n",
            "                                 (None, 120),                                                     \n",
            "                                 (None, 120)]                                                     \n",
            "                                                                                                  \n",
            " embedding_2 (Embedding)        (None, None, 120)    1213440     ['input_4[0][0]']                \n",
            "                                                                                                  \n",
            " concatenate_3 (Concatenate)    (None, 240)          0           ['bidirectional_1[0][1]',        \n",
            "                                                                  'bidirectional_1[0][3]']        \n",
            "                                                                                                  \n",
            " concatenate_4 (Concatenate)    (None, 240)          0           ['bidirectional_1[0][2]',        \n",
            "                                                                  'bidirectional_1[0][4]']        \n",
            "                                                                                                  \n",
            " lstm_3 (LSTM)                  [(None, None, 240),  346560      ['embedding_2[0][0]',            \n",
            "                                 (None, 240),                     'concatenate_3[0][0]',          \n",
            "                                 (None, 240)]                     'concatenate_4[0][0]']          \n",
            "                                                                                                  \n",
            " attention_1 (Attention)        (None, None, 240)    0           ['lstm_3[0][0]',                 \n",
            "                                                                  'bidirectional_1[0][0]']        \n",
            "                                                                                                  \n",
            " concatenate_5 (Concatenate)    (None, None, 480)    0           ['lstm_3[0][0]',                 \n",
            "                                                                  'attention_1[0][0]']            \n",
            "                                                                                                  \n",
            " dense_2 (Dense)                (None, None, 240)    115440      ['concatenate_5[0][0]']          \n",
            "                                                                                                  \n",
            " layer_normalization_1 (LayerNo  (None, None, 240)   480         ['dense_2[0][0]']                \n",
            " rmalization)                                                                                     \n",
            "                                                                                                  \n",
            " dropout_1 (Dropout)            (None, None, 240)    0           ['layer_normalization_1[0][0]']  \n",
            "                                                                                                  \n",
            " add_1 (Add)                    (None, None, 240)    0           ['lstm_3[0][0]',                 \n",
            "                                                                  'dropout_1[0][0]']              \n",
            "                                                                                                  \n",
            " dense_3 (Dense)                (None, None, 10112)  2436992     ['add_1[0][0]']                  \n",
            "                                                                                                  \n",
            "==================================================================================================\n",
            "Total params: 7,550,672\n",
            "Trainable params: 4,517,072\n",
            "Non-trainable params: 3,033,600\n",
            "__________________________________________________________________________________________________\n",
            "Epoch 1/100\n"
          ]
        },
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "2023-12-14 18:01:34.777351: W tensorflow/core/grappler/costs/op_level_cost_estimator.cc:690] Error in PredictCost() for the op: op: \"Softmax\" attr { key: \"T\" value { type: DT_FLOAT } } inputs { dtype: DT_FLOAT shape { unknown_rank: true } } device { type: \"CPU\" vendor: \"GenuineIntel\" model: \"101\" frequency: 3099 num_cores: 12 environment { key: \"cpu_instruction_set\" value: \"SSE, SSE2, SSE3\" } environment { key: \"eigen\" value: \"3.4.90\" } l1_cache_size: 32768 l2_cache_size: 262144 l3_cache_size: 12582912 memory_size: 268435456 } outputs { dtype: DT_FLOAT shape { unknown_rank: true } }\n"
          ]
        },
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "101/101 [==============================] - ETA: 0s - loss: 7.3502 - accuracy: 0.0805"
          ]
        },
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "2023-12-14 18:04:54.239938: W tensorflow/core/grappler/costs/op_level_cost_estimator.cc:690] Error in PredictCost() for the op: op: \"Softmax\" attr { key: \"T\" value { type: DT_FLOAT } } inputs { dtype: DT_FLOAT shape { unknown_rank: true } } device { type: \"CPU\" vendor: \"GenuineIntel\" model: \"101\" frequency: 3099 num_cores: 12 environment { key: \"cpu_instruction_set\" value: \"SSE, SSE2, SSE3\" } environment { key: \"eigen\" value: \"3.4.90\" } l1_cache_size: 32768 l2_cache_size: 262144 l3_cache_size: 12582912 memory_size: 268435456 } outputs { dtype: DT_FLOAT shape { unknown_rank: true } }\n"
          ]
        },
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "101/101 [==============================] - 220s 2s/step - loss: 7.3502 - accuracy: 0.0805 - val_loss: 6.9271 - val_accuracy: 0.0904\n",
            "Epoch 2/100\n",
            "101/101 [==============================] - 204s 2s/step - loss: 6.7512 - accuracy: 0.1026 - val_loss: 6.6187 - val_accuracy: 0.1215\n",
            "Epoch 3/100\n",
            "101/101 [==============================] - 195s 2s/step - loss: 6.3838 - accuracy: 0.1330 - val_loss: 6.3743 - val_accuracy: 0.1401\n",
            "Epoch 4/100\n",
            "101/101 [==============================] - 187s 2s/step - loss: 6.0591 - accuracy: 0.1478 - val_loss: 6.0350 - val_accuracy: 0.1528\n",
            "Epoch 5/100\n",
            "101/101 [==============================] - 198s 2s/step - loss: 5.7023 - accuracy: 0.1634 - val_loss: 5.7036 - val_accuracy: 0.1747\n",
            "Epoch 6/100\n",
            "101/101 [==============================] - 190s 2s/step - loss: 5.3152 - accuracy: 0.1827 - val_loss: 5.3445 - val_accuracy: 0.1899\n",
            "Epoch 7/100\n",
            "101/101 [==============================] - 186s 2s/step - loss: 4.9032 - accuracy: 0.2011 - val_loss: 4.9693 - val_accuracy: 0.2065\n",
            "Epoch 8/100\n",
            "101/101 [==============================] - 190s 2s/step - loss: 4.4588 - accuracy: 0.2325 - val_loss: 4.5702 - val_accuracy: 0.2293\n",
            "Epoch 9/100\n",
            "101/101 [==============================] - 182s 2s/step - loss: 3.9924 - accuracy: 0.2762 - val_loss: 4.1176 - val_accuracy: 0.2586\n",
            "Epoch 10/100\n",
            "101/101 [==============================] - 184s 2s/step - loss: 3.5188 - accuracy: 0.3381 - val_loss: 3.6823 - val_accuracy: 0.3104\n",
            "Epoch 11/100\n",
            "101/101 [==============================] - 183s 2s/step - loss: 3.0580 - accuracy: 0.4057 - val_loss: 3.2259 - val_accuracy: 0.3797\n",
            "Epoch 12/100\n",
            "101/101 [==============================] - 186s 2s/step - loss: 2.6074 - accuracy: 0.4835 - val_loss: 2.7746 - val_accuracy: 0.4694\n",
            "Epoch 13/100\n",
            "101/101 [==============================] - 190s 2s/step - loss: 2.2147 - accuracy: 0.5487 - val_loss: 2.3591 - val_accuracy: 0.5365\n",
            "Epoch 14/100\n",
            "101/101 [==============================] - 186s 2s/step - loss: 1.8779 - accuracy: 0.6109 - val_loss: 2.0054 - val_accuracy: 0.5978\n",
            "Epoch 15/100\n",
            "101/101 [==============================] - 188s 2s/step - loss: 1.5861 - accuracy: 0.6638 - val_loss: 1.7038 - val_accuracy: 0.6519\n",
            "Epoch 16/100\n",
            "101/101 [==============================] - 186s 2s/step - loss: 1.3516 - accuracy: 0.7071 - val_loss: 1.4292 - val_accuracy: 0.7128\n",
            "Epoch 17/100\n",
            "101/101 [==============================] - 201s 2s/step - loss: 1.1594 - accuracy: 0.7474 - val_loss: 1.2108 - val_accuracy: 0.7551\n",
            "Epoch 18/100\n",
            "101/101 [==============================] - 207s 2s/step - loss: 0.9959 - accuracy: 0.7787 - val_loss: 1.0519 - val_accuracy: 0.7855\n",
            "Epoch 19/100\n",
            "101/101 [==============================] - 205s 2s/step - loss: 0.8608 - accuracy: 0.8051 - val_loss: 0.8880 - val_accuracy: 0.8225\n",
            "Epoch 20/100\n",
            "101/101 [==============================] - 187s 2s/step - loss: 0.7370 - accuracy: 0.8307 - val_loss: 0.7752 - val_accuracy: 0.8451\n",
            "Epoch 21/100\n",
            "101/101 [==============================] - 188s 2s/step - loss: 0.6465 - accuracy: 0.8513 - val_loss: 0.6560 - val_accuracy: 0.8713\n",
            "Epoch 22/100\n",
            "101/101 [==============================] - 223s 2s/step - loss: 0.5668 - accuracy: 0.8679 - val_loss: 0.5649 - val_accuracy: 0.8971\n",
            "Epoch 23/100\n",
            "101/101 [==============================] - 210s 2s/step - loss: 0.4875 - accuracy: 0.8850 - val_loss: 0.4889 - val_accuracy: 0.9131\n",
            "Epoch 24/100\n",
            "101/101 [==============================] - 197s 2s/step - loss: 0.4302 - accuracy: 0.8988 - val_loss: 0.4344 - val_accuracy: 0.9264\n",
            "Epoch 25/100\n",
            "101/101 [==============================] - 195s 2s/step - loss: 0.3760 - accuracy: 0.9118 - val_loss: 0.3852 - val_accuracy: 0.9385\n",
            "Epoch 26/100\n",
            "101/101 [==============================] - 190s 2s/step - loss: 0.3278 - accuracy: 0.9212 - val_loss: 0.3511 - val_accuracy: 0.9469\n",
            "Epoch 27/100\n",
            "101/101 [==============================] - 215s 2s/step - loss: 0.3018 - accuracy: 0.9285 - val_loss: 0.3214 - val_accuracy: 0.9550\n",
            "Epoch 28/100\n",
            "101/101 [==============================] - 212s 2s/step - loss: 0.2698 - accuracy: 0.9351 - val_loss: 0.2999 - val_accuracy: 0.9593\n",
            "Epoch 29/100\n",
            "101/101 [==============================] - 197s 2s/step - loss: 0.2480 - accuracy: 0.9405 - val_loss: 0.2804 - val_accuracy: 0.9636\n",
            "Epoch 30/100\n",
            "101/101 [==============================] - 190s 2s/step - loss: 0.2284 - accuracy: 0.9441 - val_loss: 0.2632 - val_accuracy: 0.9700\n",
            "Epoch 31/100\n",
            "101/101 [==============================] - 190s 2s/step - loss: 0.2008 - accuracy: 0.9511 - val_loss: 0.2441 - val_accuracy: 0.9757\n",
            "Epoch 32/100\n",
            "101/101 [==============================] - 197s 2s/step - loss: 0.1830 - accuracy: 0.9552 - val_loss: 0.2339 - val_accuracy: 0.9780\n",
            "Epoch 33/100\n",
            "101/101 [==============================] - 191s 2s/step - loss: 0.1734 - accuracy: 0.9571 - val_loss: 0.2343 - val_accuracy: 0.9779\n",
            "Epoch 34/100\n",
            "101/101 [==============================] - 199s 2s/step - loss: 0.1608 - accuracy: 0.9602 - val_loss: 0.2290 - val_accuracy: 0.9789\n",
            "Epoch 35/100\n",
            "101/101 [==============================] - 197s 2s/step - loss: 0.1584 - accuracy: 0.9610 - val_loss: 0.2245 - val_accuracy: 0.9804\n",
            "Epoch 36/100\n",
            "101/101 [==============================] - 196s 2s/step - loss: 0.1427 - accuracy: 0.9646 - val_loss: 0.2190 - val_accuracy: 0.9806\n",
            "Epoch 37/100\n",
            "101/101 [==============================] - 198s 2s/step - loss: 0.1401 - accuracy: 0.9645 - val_loss: 0.2187 - val_accuracy: 0.9815\n",
            "Epoch 38/100\n",
            "101/101 [==============================] - 206s 2s/step - loss: 0.1255 - accuracy: 0.9686 - val_loss: 0.2091 - val_accuracy: 0.9839\n",
            "Epoch 39/100\n",
            "101/101 [==============================] - 211s 2s/step - loss: 0.1187 - accuracy: 0.9706 - val_loss: 0.2111 - val_accuracy: 0.9839\n",
            "Epoch 40/100\n",
            "101/101 [==============================] - 192s 2s/step - loss: 0.1069 - accuracy: 0.9736 - val_loss: 0.2137 - val_accuracy: 0.9845\n",
            "Epoch 41/100\n",
            "101/101 [==============================] - 195s 2s/step - loss: 0.1050 - accuracy: 0.9730 - val_loss: 0.2038 - val_accuracy: 0.9857\n",
            "Epoch 42/100\n",
            "101/101 [==============================] - 190s 2s/step - loss: 0.1027 - accuracy: 0.9739 - val_loss: 0.2068 - val_accuracy: 0.9853\n",
            "Epoch 43/100\n",
            "101/101 [==============================] - 188s 2s/step - loss: 0.0929 - accuracy: 0.9763 - val_loss: 0.2112 - val_accuracy: 0.9862\n",
            "Epoch 44/100\n",
            "101/101 [==============================] - 188s 2s/step - loss: 0.0880 - accuracy: 0.9780 - val_loss: 0.1996 - val_accuracy: 0.9869\n",
            "Epoch 45/100\n",
            "101/101 [==============================] - 190s 2s/step - loss: 0.0914 - accuracy: 0.9775 - val_loss: 0.2204 - val_accuracy: 0.9821\n",
            "Epoch 46/100\n",
            "101/101 [==============================] - 189s 2s/step - loss: 0.0982 - accuracy: 0.9759 - val_loss: 0.2018 - val_accuracy: 0.9865\n",
            "Epoch 47/100\n",
            "101/101 [==============================] - 198s 2s/step - loss: 0.0862 - accuracy: 0.9773 - val_loss: 0.2071 - val_accuracy: 0.9868\n"
          ]
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "# Define the EarlyStopping callback\n",
        "early_stopping = EarlyStopping(monitor='val_loss', patience=3, restore_best_weights=True)\n",
        "\n",
        "# Compile the model with the EarlyStopping callback\n",
        "model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])\n",
        "\n",
        "# Print the model summary\n",
        "model.summary()\n",
        "\n",
        "# Fit the model with early stopping\n",
        "history = model.fit([encoder_input_data, decoder_input_data], decoder_target_data,\n",
        "                    batch_size=32,\n",
        "                    epochs=100,\n",
        "                    validation_split=0.2,\n",
        "                    callbacks=[early_stopping])\n",
        "\n",
        "# Plot training and validation accuracy\n",
        "plt.plot(history.history['accuracy'], label='Training Accuracy')\n",
        "plt.plot(history.history['val_accuracy'], label='Validation Accuracy')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.title('Training and Validation Accuracy Over Epochs')\n",
        "plt.legend()\n",
        "plt.show()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "4syuqY0wgP_0"
      },
      "outputs": [],
      "source": [
        "# Save the model\n",
        "model.save('model1.h5')\n",
        "\n",
        "# Save the model in the native Keras format\n",
        "model.save('model1.keras')\n",
        "\n",
        "# Optionally, saving model's weights separately\n",
        "model.save_weights('model1_weights.h5')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Yh9ZDtSMgP_0"
      },
      "source": [
        "Training the Model(2)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "cwfE2-wogP_0",
        "outputId": "0353b2a8-2264-4f96-ffcd-ceb1f6bd8259"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Model: \"model_6\"\n",
            "__________________________________________________________________________________________________\n",
            " Layer (type)                   Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            " input_15 (InputLayer)          [(None, 140)]        0           []                               \n",
            "                                                                                                  \n",
            " embedding_7 (Embedding)        (None, 140, 300)     3033600     ['input_15[0][0]']               \n",
            "                                                                                                  \n",
            " input_16 (InputLayer)          [(None, None)]       0           []                               \n",
            "                                                                                                  \n",
            " bidirectional_7 (Bidirectional  [(None, 140, 180),  281520      ['embedding_7[1][0]']            \n",
            " )                               (None, 90),                                                      \n",
            "                                 (None, 90),                                                      \n",
            "                                 (None, 90),                                                      \n",
            "                                 (None, 90)]                                                      \n",
            "                                                                                                  \n",
            " embedding_9 (Embedding)        (None, None, 90)     910080      ['input_16[0][0]']               \n",
            "                                                                                                  \n",
            " concatenate_21 (Concatenate)   (None, 180)          0           ['bidirectional_7[0][1]',        \n",
            "                                                                  'bidirectional_7[0][3]']        \n",
            "                                                                                                  \n",
            " concatenate_22 (Concatenate)   (None, 180)          0           ['bidirectional_7[0][2]',        \n",
            "                                                                  'bidirectional_7[0][4]']        \n",
            "                                                                                                  \n",
            " lstm_15 (LSTM)                 [(None, None, 180),  195120      ['embedding_9[0][0]',            \n",
            "                                 (None, 180),                     'concatenate_21[0][0]',         \n",
            "                                 (None, 180)]                     'concatenate_22[0][0]']         \n",
            "                                                                                                  \n",
            " attention_7 (Attention)        (None, None, 180)    0           ['lstm_15[0][0]',                \n",
            "                                                                  'bidirectional_7[0][0]']        \n",
            "                                                                                                  \n",
            " concatenate_23 (Concatenate)   (None, None, 360)    0           ['lstm_15[0][0]',                \n",
            "                                                                  'attention_7[0][0]']            \n",
            "                                                                                                  \n",
            " dense_13 (Dense)               (None, None, 90)     32490       ['concatenate_23[0][0]']         \n",
            "                                                                                                  \n",
            " dropout_7 (Dropout)            (None, None, 90)     0           ['dense_13[0][0]']               \n",
            "                                                                                                  \n",
            " dense_14 (Dense)               (None, None, 10112)  920192      ['dropout_7[0][0]']              \n",
            "                                                                                                  \n",
            "==================================================================================================\n",
            "Total params: 5,373,002\n",
            "Trainable params: 2,339,402\n",
            "Non-trainable params: 3,033,600\n",
            "__________________________________________________________________________________________________\n",
            "Epoch 1/50\n",
            "101/101 [==============================] - 287s 3s/step - loss: 7.5407 - accuracy: 0.0707 - val_loss: 6.9961 - val_accuracy: 0.0854 - lr: 0.0010\n",
            "Epoch 2/50\n",
            "101/101 [==============================] - 254s 3s/step - loss: 6.8094 - accuracy: 0.0931 - val_loss: 6.8019 - val_accuracy: 0.1011 - lr: 9.0484e-04\n",
            "Epoch 3/50\n",
            "101/101 [==============================] - 242s 2s/step - loss: 6.5971 - accuracy: 0.1087 - val_loss: 6.6290 - val_accuracy: 0.1166 - lr: 8.1873e-04\n",
            "Epoch 4/50\n",
            "101/101 [==============================] - 240s 2s/step - loss: 6.3969 - accuracy: 0.1247 - val_loss: 6.4575 - val_accuracy: 0.1297 - lr: 7.4082e-04\n",
            "Epoch 5/50\n",
            "101/101 [==============================] - 240s 2s/step - loss: 6.2430 - accuracy: 0.1332 - val_loss: 6.3386 - val_accuracy: 0.1374 - lr: 6.7032e-04\n",
            "Epoch 6/50\n",
            "101/101 [==============================] - 239s 2s/step - loss: 6.1079 - accuracy: 0.1401 - val_loss: 6.2152 - val_accuracy: 0.1465 - lr: 6.0653e-04\n",
            "Epoch 7/50\n",
            "101/101 [==============================] - 241s 2s/step - loss: 5.9778 - accuracy: 0.1479 - val_loss: 6.1161 - val_accuracy: 0.1523 - lr: 5.4881e-04\n",
            "Epoch 8/50\n",
            "101/101 [==============================] - 236s 2s/step - loss: 5.8520 - accuracy: 0.1537 - val_loss: 6.0080 - val_accuracy: 0.1599 - lr: 4.9659e-04\n",
            "Epoch 9/50\n",
            "101/101 [==============================] - 239s 2s/step - loss: 5.7186 - accuracy: 0.1624 - val_loss: 5.9056 - val_accuracy: 0.1682 - lr: 4.4933e-04\n",
            "Epoch 10/50\n",
            "101/101 [==============================] - 237s 2s/step - loss: 5.5915 - accuracy: 0.1688 - val_loss: 5.8062 - val_accuracy: 0.1735 - lr: 4.0657e-04\n",
            "Epoch 11/50\n",
            "101/101 [==============================] - 316s 3s/step - loss: 5.4792 - accuracy: 0.1758 - val_loss: 5.7167 - val_accuracy: 0.1803 - lr: 3.6788e-04\n",
            "Epoch 12/50\n",
            "101/101 [==============================] - 351s 3s/step - loss: 5.3707 - accuracy: 0.1813 - val_loss: 5.6318 - val_accuracy: 0.1829 - lr: 3.3287e-04\n",
            "Epoch 13/50\n",
            "101/101 [==============================] - 306s 3s/step - loss: 5.2734 - accuracy: 0.1865 - val_loss: 5.5556 - val_accuracy: 0.1868 - lr: 3.0119e-04\n",
            "Epoch 14/50\n",
            "101/101 [==============================] - 283s 3s/step - loss: 5.1788 - accuracy: 0.1904 - val_loss: 5.4854 - val_accuracy: 0.1896 - lr: 2.7253e-04\n",
            "Epoch 15/50\n",
            "101/101 [==============================] - 227s 2s/step - loss: 5.0861 - accuracy: 0.1938 - val_loss: 5.4078 - val_accuracy: 0.1953 - lr: 2.4660e-04\n",
            "Epoch 16/50\n",
            "101/101 [==============================] - 252s 2s/step - loss: 5.0020 - accuracy: 0.1980 - val_loss: 5.3459 - val_accuracy: 0.1982 - lr: 2.2313e-04\n",
            "Epoch 17/50\n",
            "101/101 [==============================] - 228s 2s/step - loss: 4.9297 - accuracy: 0.2009 - val_loss: 5.2819 - val_accuracy: 0.1986 - lr: 2.0190e-04\n",
            "Epoch 18/50\n",
            "101/101 [==============================] - 228s 2s/step - loss: 4.8610 - accuracy: 0.2058 - val_loss: 5.2330 - val_accuracy: 0.1997 - lr: 1.8268e-04\n",
            "Epoch 19/50\n",
            "101/101 [==============================] - 223s 2s/step - loss: 4.8034 - accuracy: 0.2072 - val_loss: 5.1830 - val_accuracy: 0.2039 - lr: 1.6530e-04\n",
            "Epoch 20/50\n",
            "101/101 [==============================] - 222s 2s/step - loss: 4.7529 - accuracy: 0.2090 - val_loss: 5.1427 - val_accuracy: 0.2036 - lr: 1.4957e-04\n",
            "Epoch 21/50\n",
            "101/101 [==============================] - 223s 2s/step - loss: 4.7082 - accuracy: 0.2120 - val_loss: 5.1052 - val_accuracy: 0.2065 - lr: 1.3534e-04\n",
            "Epoch 22/50\n",
            "101/101 [==============================] - 230s 2s/step - loss: 4.6660 - accuracy: 0.2152 - val_loss: 5.0701 - val_accuracy: 0.2082 - lr: 1.2246e-04\n",
            "Epoch 23/50\n",
            "101/101 [==============================] - 260s 3s/step - loss: 4.6288 - accuracy: 0.2145 - val_loss: 5.0405 - val_accuracy: 0.2083 - lr: 1.1080e-04\n",
            "Epoch 24/50\n",
            "101/101 [==============================] - 259s 3s/step - loss: 4.5984 - accuracy: 0.2171 - val_loss: 5.0095 - val_accuracy: 0.2109 - lr: 1.0026e-04\n",
            "Epoch 25/50\n",
            "101/101 [==============================] - 238s 2s/step - loss: 4.5628 - accuracy: 0.2177 - val_loss: 4.9885 - val_accuracy: 0.2121 - lr: 9.0718e-05\n",
            "Epoch 26/50\n",
            "101/101 [==============================] - 243s 2s/step - loss: 4.5376 - accuracy: 0.2196 - val_loss: 4.9677 - val_accuracy: 0.2124 - lr: 8.2085e-05\n",
            "Epoch 27/50\n",
            "101/101 [==============================] - 237s 2s/step - loss: 4.5143 - accuracy: 0.2218 - val_loss: 4.9451 - val_accuracy: 0.2142 - lr: 7.4274e-05\n",
            "Epoch 28/50\n",
            "101/101 [==============================] - 230s 2s/step - loss: 4.4896 - accuracy: 0.2221 - val_loss: 4.9250 - val_accuracy: 0.2142 - lr: 6.7206e-05\n",
            "Epoch 29/50\n",
            "101/101 [==============================] - 240s 2s/step - loss: 4.4697 - accuracy: 0.2250 - val_loss: 4.9100 - val_accuracy: 0.2154 - lr: 6.0810e-05\n",
            "Epoch 30/50\n",
            "101/101 [==============================] - 249s 2s/step - loss: 4.4462 - accuracy: 0.2260 - val_loss: 4.8933 - val_accuracy: 0.2169 - lr: 5.5023e-05\n",
            "Epoch 31/50\n",
            "101/101 [==============================] - 240s 2s/step - loss: 4.4340 - accuracy: 0.2259 - val_loss: 4.8805 - val_accuracy: 0.2175 - lr: 4.9787e-05\n",
            "Epoch 32/50\n",
            "101/101 [==============================] - 233s 2s/step - loss: 4.4194 - accuracy: 0.2292 - val_loss: 4.8655 - val_accuracy: 0.2189 - lr: 4.5049e-05\n",
            "Epoch 33/50\n",
            "101/101 [==============================] - 230s 2s/step - loss: 4.4094 - accuracy: 0.2284 - val_loss: 4.8581 - val_accuracy: 0.2178 - lr: 4.0762e-05\n",
            "Epoch 34/50\n",
            "101/101 [==============================] - 229s 2s/step - loss: 4.3984 - accuracy: 0.2282 - val_loss: 4.8441 - val_accuracy: 0.2189 - lr: 3.6883e-05\n",
            "Epoch 35/50\n",
            "101/101 [==============================] - 229s 2s/step - loss: 4.3837 - accuracy: 0.2286 - val_loss: 4.8352 - val_accuracy: 0.2201 - lr: 3.3373e-05\n",
            "Epoch 36/50\n",
            "101/101 [==============================] - 237s 2s/step - loss: 4.3757 - accuracy: 0.2305 - val_loss: 4.8295 - val_accuracy: 0.2195 - lr: 3.0197e-05\n",
            "Epoch 37/50\n",
            "101/101 [==============================] - 244s 2s/step - loss: 4.3725 - accuracy: 0.2298 - val_loss: 4.8204 - val_accuracy: 0.2202 - lr: 2.7324e-05\n",
            "Epoch 38/50\n",
            "101/101 [==============================] - 230s 2s/step - loss: 4.3596 - accuracy: 0.2312 - val_loss: 4.8131 - val_accuracy: 0.2203 - lr: 2.4724e-05\n",
            "Epoch 39/50\n",
            "101/101 [==============================] - 230s 2s/step - loss: 4.3482 - accuracy: 0.2322 - val_loss: 4.8060 - val_accuracy: 0.2204 - lr: 2.2371e-05\n",
            "Epoch 40/50\n",
            "101/101 [==============================] - 229s 2s/step - loss: 4.3451 - accuracy: 0.2339 - val_loss: 4.8032 - val_accuracy: 0.2206 - lr: 2.0242e-05\n",
            "Epoch 41/50\n",
            "101/101 [==============================] - 231s 2s/step - loss: 4.3351 - accuracy: 0.2333 - val_loss: 4.7972 - val_accuracy: 0.2207 - lr: 1.8316e-05\n",
            "Epoch 42/50\n",
            "101/101 [==============================] - 232s 2s/step - loss: 4.3284 - accuracy: 0.2337 - val_loss: 4.7920 - val_accuracy: 0.2209 - lr: 1.6573e-05\n",
            "Epoch 43/50\n",
            "101/101 [==============================] - 229s 2s/step - loss: 4.3272 - accuracy: 0.2332 - val_loss: 4.7872 - val_accuracy: 0.2208 - lr: 1.4996e-05\n",
            "Epoch 44/50\n",
            "101/101 [==============================] - 229s 2s/step - loss: 4.3226 - accuracy: 0.2336 - val_loss: 4.7837 - val_accuracy: 0.2205 - lr: 1.3569e-05\n",
            "Epoch 45/50\n",
            "101/101 [==============================] - 231s 2s/step - loss: 4.3205 - accuracy: 0.2330 - val_loss: 4.7793 - val_accuracy: 0.2210 - lr: 1.2277e-05\n",
            "Epoch 46/50\n",
            "101/101 [==============================] - 230s 2s/step - loss: 4.3176 - accuracy: 0.2326 - val_loss: 4.7772 - val_accuracy: 0.2213 - lr: 1.1109e-05\n",
            "Epoch 47/50\n",
            "101/101 [==============================] - 228s 2s/step - loss: 4.3143 - accuracy: 0.2350 - val_loss: 4.7748 - val_accuracy: 0.2210 - lr: 1.0052e-05\n",
            "Epoch 48/50\n",
            "101/101 [==============================] - 237s 2s/step - loss: 4.3107 - accuracy: 0.2341 - val_loss: 4.7722 - val_accuracy: 0.2210 - lr: 9.0953e-06\n",
            "Epoch 49/50\n",
            "101/101 [==============================] - 233s 2s/step - loss: 4.3042 - accuracy: 0.2348 - val_loss: 4.7705 - val_accuracy: 0.2213 - lr: 8.2297e-06\n",
            "Epoch 50/50\n",
            "101/101 [==============================] - 235s 2s/step - loss: 4.3016 - accuracy: 0.2360 - val_loss: 4.7674 - val_accuracy: 0.2219 - lr: 7.4466e-06\n"
          ]
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "import tensorflow as tf\n",
        "from tensorflow.keras.layers import Input, LSTM, Dense, Embedding, Bidirectional, Concatenate, Attention\n",
        "from tensorflow.keras.models import Model\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "from tensorflow.keras.layers import Dropout\n",
        "\n",
        "# Convert data to tensors\n",
        "encoder_input_data = tf.convert_to_tensor(encoder_input_data, dtype=tf.float32)\n",
        "decoder_input_data = tf.convert_to_tensor(decoder_input_data, dtype=tf.float32)\n",
        "decoder_target_data = tf.convert_to_tensor(decoder_target_data, dtype=tf.float32)\n",
        "\n",
        "# Define the parameters\n",
        "latent_dim = 90\n",
        "num_encoder_tokens = len(tokenizer.word_index) + 1  # Adjust according to your data\n",
        "num_decoder_tokens = len(tokenizer.word_index) + 1  # Adjust according to your data\n",
        "\n",
        "# Define the encoder\n",
        "encoder_inputs = Input(shape=(max_sequence_length,))\n",
        "encoder_embedding = encoder_embedding_layer(encoder_inputs)\n",
        "encoder_bi_lstm = Bidirectional(LSTM(latent_dim, return_sequences=True, return_state=True, dropout=0.2, recurrent_dropout=0.2))\n",
        "encoder_outputs, forward_h, forward_c, backward_h, backward_c = encoder_bi_lstm(encoder_embedding)\n",
        "state_h = Concatenate()([forward_h, backward_h])\n",
        "state_c = Concatenate()([forward_c, backward_c])\n",
        "\n",
        "# Define the decoder\n",
        "decoder_inputs = Input(shape=(None,))\n",
        "decoder_embedding = Embedding(num_decoder_tokens, latent_dim, mask_zero=True)(decoder_inputs)\n",
        "decoder_lstm = LSTM(latent_dim*2, return_sequences=True, return_state=True, dropout=0.2, recurrent_dropout=0.2)\n",
        "decoder_outputs, _, _ = decoder_lstm(decoder_embedding, initial_state=[state_h, state_c])\n",
        "attention = Attention()([decoder_outputs, encoder_outputs])\n",
        "decoder_concat = tf.keras.layers.concatenate([decoder_outputs, attention], axis=-1)\n",
        "\n",
        "# Additional Dense Layer\n",
        "decoder_dense_2 = Dense(latent_dim, activation='relu')\n",
        "decoder_outputs_2 = decoder_dense_2(decoder_concat)\n",
        "decoder_dropout = Dropout(0.2)(decoder_outputs_2)  # Apply Dropout directly to the output\n",
        "\n",
        "# Final Dense Layer\n",
        "decoder_dense = Dense(num_decoder_tokens, activation='softmax')\n",
        "decoder_outputs = decoder_dense(decoder_dropout)  # Use the output after applying Dropout\n",
        "\n",
        "\n",
        "# Define the model\n",
        "model = Model([encoder_inputs, decoder_inputs], decoder_outputs)\n",
        "\n",
        "# Compile the model\n",
        "model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])\n",
        "\n",
        "# Print the model summary\n",
        "model.summary()\n",
        "\n",
        "# Learning Rate Schedule\n",
        "def lr_schedule(epoch):\n",
        "    return 0.001 * np.exp(-epoch / 10)\n",
        "\n",
        "lr_callback = tf.keras.callbacks.LearningRateScheduler(lr_schedule)\n",
        "\n",
        "# Train the model\n",
        "history = model.fit([encoder_input_data, decoder_input_data], decoder_target_data,\n",
        "                    batch_size=32,\n",
        "                    epochs=50,\n",
        "                    validation_split=0.2,callbacks=[lr_callback]\n",
        "                    )\n",
        "\n",
        "# Plot training and validation accuracy\n",
        "plt.plot(history.history['accuracy'], label='Training Accuracy')\n",
        "plt.plot(history.history['val_accuracy'], label='Validation Accuracy')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.title('Training and Validation Accuracy Over Epochs')\n",
        "plt.legend()\n",
        "plt.show()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Hib9JessgP_0"
      },
      "outputs": [],
      "source": [
        "# Save the model\n",
        "model.save('model2.h5')\n",
        "\n",
        "# Save the model in the native Keras format\n",
        "model.save('model2.keras')\n",
        "\n",
        "# Optionally, saving model's weights separately\n",
        "model.save_weights('model2_weights.h5')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SeXr5KfzgP_1"
      },
      "source": [
        "Model -3"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "CICiDV0HgP_1",
        "outputId": "8468f64b-d912-4dee-aeeb-f793d9d49dae"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Model: \"model\"\n",
            "__________________________________________________________________________________________________\n",
            " Layer (type)                   Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            " input_1 (InputLayer)           [(None, 140)]        0           []                               \n",
            "                                                                                                  \n",
            " embedding (Embedding)          (None, 140, 300)     3033600     ['input_1[0][0]']                \n",
            "                                                                                                  \n",
            " input_2 (InputLayer)           [(None, None)]       0           []                               \n",
            "                                                                                                  \n",
            " bidirectional (Bidirectional)  [(None, 140, 180),   281520      ['embedding[0][0]']              \n",
            "                                 (None, 90),                                                      \n",
            "                                 (None, 90),                                                      \n",
            "                                 (None, 90),                                                      \n",
            "                                 (None, 90)]                                                      \n",
            "                                                                                                  \n",
            " embedding_1 (Embedding)        (None, None, 90)     910080      ['input_2[0][0]']                \n",
            "                                                                                                  \n",
            " concatenate (Concatenate)      (None, 180)          0           ['bidirectional[0][1]',          \n",
            "                                                                  'bidirectional[0][3]']          \n",
            "                                                                                                  \n",
            " concatenate_1 (Concatenate)    (None, 180)          0           ['bidirectional[0][2]',          \n",
            "                                                                  'bidirectional[0][4]']          \n",
            "                                                                                                  \n",
            " lstm_1 (LSTM)                  [(None, None, 180),  195120      ['embedding_1[0][0]',            \n",
            "                                 (None, 180),                     'concatenate[0][0]',            \n",
            "                                 (None, 180)]                     'concatenate_1[0][0]']          \n",
            "                                                                                                  \n",
            " attention (Attention)          (None, None, 180)    0           ['lstm_1[0][0]',                 \n",
            "                                                                  'bidirectional[0][0]']          \n",
            "                                                                                                  \n",
            " concatenate_2 (Concatenate)    (None, None, 360)    0           ['lstm_1[0][0]',                 \n",
            "                                                                  'attention[0][0]']              \n",
            "                                                                                                  \n",
            " dense (Dense)                  (None, None, 90)     32490       ['concatenate_2[0][0]']          \n",
            "                                                                                                  \n",
            " dropout (Dropout)              (None, None, 90)     0           ['dense[0][0]']                  \n",
            "                                                                                                  \n",
            " dense_1 (Dense)                (None, None, 10112)  920192      ['dropout[0][0]']                \n",
            "                                                                                                  \n",
            "==================================================================================================\n",
            "Total params: 5,373,002\n",
            "Trainable params: 2,339,402\n",
            "Non-trainable params: 3,033,600\n",
            "__________________________________________________________________________________________________\n",
            "Epoch 1/25\n",
            "88/88 [==============================] - 224s 2s/step - loss: 7.6017 - accuracy: 0.0688 - val_loss: 7.0129 - val_accuracy: 0.0876\n",
            "Epoch 2/25\n",
            "88/88 [==============================] - 212s 2s/step - loss: 6.8616 - accuracy: 0.0867 - val_loss: 6.7952 - val_accuracy: 0.0936\n",
            "Epoch 3/25\n",
            "88/88 [==============================] - 203s 2s/step - loss: 6.6587 - accuracy: 0.0962 - val_loss: 6.6254 - val_accuracy: 0.1099\n",
            "Epoch 4/25\n",
            "88/88 [==============================] - 203s 2s/step - loss: 6.4487 - accuracy: 0.1218 - val_loss: 6.4237 - val_accuracy: 0.1321\n",
            "Epoch 5/25\n",
            "88/88 [==============================] - 203s 2s/step - loss: 6.2697 - accuracy: 0.1341 - val_loss: 6.2782 - val_accuracy: 0.1389\n",
            "Epoch 6/25\n",
            "88/88 [==============================] - 205s 2s/step - loss: 6.1002 - accuracy: 0.1443 - val_loss: 6.0968 - val_accuracy: 0.1511\n",
            "Epoch 7/25\n",
            "88/88 [==============================] - 204s 2s/step - loss: 5.9205 - accuracy: 0.1561 - val_loss: 5.9334 - val_accuracy: 0.1652\n",
            "Epoch 8/25\n",
            "88/88 [==============================] - 212s 2s/step - loss: 5.7462 - accuracy: 0.1683 - val_loss: 5.7900 - val_accuracy: 0.1778\n",
            "Epoch 9/25\n",
            "88/88 [==============================] - 220s 3s/step - loss: 5.5880 - accuracy: 0.1795 - val_loss: 5.6349 - val_accuracy: 0.1900\n",
            "Epoch 10/25\n",
            "88/88 [==============================] - 214s 2s/step - loss: 5.4368 - accuracy: 0.1895 - val_loss: 5.4996 - val_accuracy: 0.1963\n",
            "Epoch 11/25\n",
            "88/88 [==============================] - 221s 3s/step - loss: 5.2921 - accuracy: 0.1982 - val_loss: 5.3595 - val_accuracy: 0.2028\n",
            "Epoch 12/25\n",
            "88/88 [==============================] - 220s 3s/step - loss: 5.1436 - accuracy: 0.2044 - val_loss: 5.2062 - val_accuracy: 0.2096\n",
            "Epoch 13/25\n",
            "88/88 [==============================] - 217s 2s/step - loss: 5.0044 - accuracy: 0.2117 - val_loss: 5.0693 - val_accuracy: 0.2160\n",
            "Epoch 14/25\n",
            "88/88 [==============================] - 212s 2s/step - loss: 4.8598 - accuracy: 0.2169 - val_loss: 4.9293 - val_accuracy: 0.2225\n",
            "Epoch 15/25\n",
            "88/88 [==============================] - 216s 2s/step - loss: 4.7102 - accuracy: 0.2236 - val_loss: 4.7752 - val_accuracy: 0.2277\n",
            "Epoch 16/25\n",
            "88/88 [==============================] - 214s 2s/step - loss: 4.5642 - accuracy: 0.2316 - val_loss: 4.6086 - val_accuracy: 0.2344\n",
            "Epoch 17/25\n",
            "88/88 [==============================] - 209s 2s/step - loss: 4.4198 - accuracy: 0.2395 - val_loss: 4.4596 - val_accuracy: 0.2405\n",
            "Epoch 18/25\n",
            "88/88 [==============================] - 217s 2s/step - loss: 4.2763 - accuracy: 0.2501 - val_loss: 4.3079 - val_accuracy: 0.2491\n",
            "Epoch 19/25\n",
            "88/88 [==============================] - 215s 2s/step - loss: 4.1146 - accuracy: 0.2606 - val_loss: 4.1526 - val_accuracy: 0.2547\n",
            "Epoch 20/25\n",
            "88/88 [==============================] - 216s 2s/step - loss: 3.9882 - accuracy: 0.2725 - val_loss: 4.0004 - val_accuracy: 0.2693\n",
            "Epoch 21/25\n",
            "88/88 [==============================] - 205s 2s/step - loss: 3.8359 - accuracy: 0.2862 - val_loss: 3.8660 - val_accuracy: 0.2832\n",
            "Epoch 22/25\n",
            "88/88 [==============================] - 203s 2s/step - loss: 3.6974 - accuracy: 0.3009 - val_loss: 3.7008 - val_accuracy: 0.2970\n",
            "Epoch 23/25\n",
            "88/88 [==============================] - 217s 2s/step - loss: 3.5727 - accuracy: 0.3144 - val_loss: 3.5524 - val_accuracy: 0.3118\n",
            "Epoch 24/25\n",
            "88/88 [==============================] - 213s 2s/step - loss: 3.4401 - accuracy: 0.3296 - val_loss: 3.4218 - val_accuracy: 0.3316\n",
            "Epoch 25/25\n",
            "88/88 [==============================] - 213s 2s/step - loss: 3.3213 - accuracy: 0.3467 - val_loss: 3.2815 - val_accuracy: 0.3512\n"
          ]
        },
        {
          "data": {
            "image/png": "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",
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ]
          },
          "metadata": {},
          "output_type": "display_data"
        }
      ],
      "source": [
        "import tensorflow as tf\n",
        "from tensorflow.keras.layers import Input, LSTM, Dense, Embedding, Bidirectional, Concatenate, Attention\n",
        "from tensorflow.keras.models import Model\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "from tensorflow.keras.layers import Dropout\n",
        "\n",
        "# Convert data to tensors\n",
        "encoder_input_data = tf.convert_to_tensor(encoder_input_data, dtype=tf.float32)\n",
        "decoder_input_data = tf.convert_to_tensor(decoder_input_data, dtype=tf.float32)\n",
        "decoder_target_data = tf.convert_to_tensor(decoder_target_data, dtype=tf.float32)\n",
        "\n",
        "# Define the parameters\n",
        "latent_dim = 90\n",
        "num_encoder_tokens = len(tokenizer.word_index) + 1  # Adjust according to your data\n",
        "num_decoder_tokens = len(tokenizer.word_index) + 1  # Adjust according to your data\n",
        "\n",
        "# Define the encoder\n",
        "encoder_inputs = Input(shape=(max_sequence_length,))\n",
        "encoder_embedding = encoder_embedding_layer(encoder_inputs)\n",
        "encoder_bi_lstm = Bidirectional(LSTM(latent_dim, return_sequences=True, return_state=True, dropout=0.2, recurrent_dropout=0.2))\n",
        "encoder_outputs, forward_h, forward_c, backward_h, backward_c = encoder_bi_lstm(encoder_embedding)\n",
        "state_h = Concatenate()([forward_h, backward_h])\n",
        "state_c = Concatenate()([forward_c, backward_c])\n",
        "\n",
        "# Define the decoder\n",
        "decoder_inputs = Input(shape=(None,))\n",
        "decoder_embedding = Embedding(num_decoder_tokens, latent_dim, mask_zero=True)(decoder_inputs)\n",
        "decoder_lstm = LSTM(latent_dim*2, return_sequences=True, return_state=True, dropout=0.2, recurrent_dropout=0.2)\n",
        "decoder_outputs, _, _ = decoder_lstm(decoder_embedding, initial_state=[state_h, state_c])\n",
        "attention = Attention()([decoder_outputs, encoder_outputs])\n",
        "decoder_concat = tf.keras.layers.concatenate([decoder_outputs, attention], axis=-1)\n",
        "\n",
        "# Additional Dense Layer\n",
        "decoder_dense_2 = Dense(latent_dim, activation='relu')\n",
        "decoder_outputs_2 = decoder_dense_2(decoder_concat)\n",
        "decoder_dropout = Dropout(0.2)(decoder_outputs_2)  # Apply Dropout directly to the output\n",
        "\n",
        "# Final Dense Layer\n",
        "decoder_dense = Dense(num_decoder_tokens, activation='softmax')\n",
        "decoder_outputs = decoder_dense(decoder_dropout)  # Use the output after applying Dropout\n",
        "\n",
        "\n",
        "# Define the model\n",
        "model = Model([encoder_inputs, decoder_inputs], decoder_outputs)\n",
        "\n",
        "# Compile the model\n",
        "model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])\n",
        "\n",
        "# Print the model summary\n",
        "model.summary()\n",
        "\n",
        "# Learning Rate Schedule\n",
        "#def lr_schedule(epoch):\n",
        "#    return 0.001 * np.exp(-epoch / 10)\n",
        "\n",
        "#lr_callback = tf.keras.callbacks.LearningRateScheduler(lr_schedule)\n",
        "#callbacks=[lr_callback]\n",
        "# Train the model\n",
        "history = model.fit([encoder_input_data, decoder_input_data], decoder_target_data,\n",
        "                    batch_size=32,\n",
        "                    epochs=25,\n",
        "                    validation_split=0.3\n",
        "                    )\n",
        "\n",
        "# Plot training and validation accuracy\n",
        "plt.plot(history.history['accuracy'], label='Training Accuracy')\n",
        "plt.plot(history.history['val_accuracy'], label='Validation Accuracy')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.title('Training and Validation Accuracy Over Epochs')\n",
        "plt.legend()\n",
        "plt.show()"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qZ5l5mZegP_1"
      },
      "source": [
        "Testing (On Model-1)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "eEDCQ44IgP_1"
      },
      "source": [
        "Loading Test Data"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "MvH7SLcSgP_1",
        "outputId": "c6dddf1f-1b57-4690-cbf5-b58aa5cd7842"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "Arrays loaded from test300+150.txt\n"
          ]
        }
      ],
      "source": [
        "# Specify the file path from where to load the arrays\n",
        "file_path = \"test300+150.txt\"\n",
        "\n",
        "# Initialize empty arrays to store loaded data\n",
        "test_input_texts = []\n",
        "test_target_texts = []\n",
        "\n",
        "# Open the file in read mode and load the arrays from the file\n",
        "with open(file_path, \"r\") as file:\n",
        "    for line in file:\n",
        "        # Split each line into input and target using the tab (\"\\t\") as a separator\n",
        "        input_text, target_text = line.strip().split(\"\\t\")\n",
        "\n",
        "        # Append the data to the respective arrays\n",
        "        test_input_texts.append(input_text)\n",
        "        test_target_texts.append(target_text)\n",
        "\n",
        "# Print a message indicating that arrays are loaded\n",
        "print(f\"Arrays loaded from {file_path}\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "TH9D2eIEgP_1"
      },
      "outputs": [],
      "source": [
        "pip install rouge"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "GQyCkxh-gP_1"
      },
      "source": [
        "Evaluation on entire test data"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "kSAXQMnCgP_1",
        "outputId": "2b907504-d4a8-4052-94a9-ef7601806a84"
      },
      "outputs": [
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "2023-12-14 22:21:04.993376: W tensorflow/core/grappler/costs/op_level_cost_estimator.cc:690] Error in PredictCost() for the op: op: \"Softmax\" attr { key: \"T\" value { type: DT_FLOAT } } inputs { dtype: DT_FLOAT shape { unknown_rank: true } } device { type: \"CPU\" vendor: \"GenuineIntel\" model: \"101\" frequency: 3099 num_cores: 12 environment { key: \"cpu_instruction_set\" value: \"SSE, SSE2, SSE3\" } environment { key: \"eigen\" value: \"3.4.90\" } l1_cache_size: 32768 l2_cache_size: 262144 l3_cache_size: 12582912 memory_size: 268435456 } outputs { dtype: DT_FLOAT shape { unknown_rank: true } }\n"
          ]
        },
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "15/15 [==============================] - 5s 193ms/step\n",
            "BLEU Score: 16.07%\n",
            "BLEU Score with Smoothing: 16.07%\n",
            "ROUGE-N Score: 0.08%\n",
            "ROUGE-L Score: 0.00%\n"
          ]
        }
      ],
      "source": [
        "import tensorflow_datasets as tfds\n",
        "import tensorflow as tf\n",
        "from tensorflow.keras.preprocessing.text import Tokenizer\n",
        "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
        "import numpy as np\n",
        "import pickle\n",
        "from nltk.translate.bleu_score import corpus_bleu, SmoothingFunction\n",
        "from nltk.translate.bleu_score import sentence_bleu\n",
        "\n",
        "# Load the tokenizer from the saved file\n",
        "with open('tokenizer.pickle', 'rb') as handle:\n",
        "    tokenizer = pickle.load(handle)\n",
        "\n",
        "# Tokenize the test data\n",
        "test_encoder_sequences = tokenizer.texts_to_sequences(test_input_texts)\n",
        "test_decoder_sequences = tokenizer.texts_to_sequences(test_target_texts)\n",
        "\n",
        "# Pad the sequences\n",
        "max_seq_length = max(max(len(seq) for seq in test_encoder_sequences), max(len(seq) for seq in test_decoder_sequences))\n",
        "test_encoder_input_data = pad_sequences(test_encoder_sequences, maxlen=max_seq_length, padding='post')\n",
        "test_decoder_input_data = pad_sequences(test_decoder_sequences, maxlen=max_seq_length, padding='post')\n",
        "\n",
        "# Convert data to tensors\n",
        "test_encoder_input_data = tf.convert_to_tensor(test_encoder_input_data, dtype=tf.int32)\n",
        "test_decoder_input_data = tf.convert_to_tensor(test_decoder_input_data, dtype=tf.int32)\n",
        "\n",
        "# Generate predictions using the trained model\n",
        "predictions = model.predict([test_encoder_input_data, test_decoder_input_data])\n",
        "\n",
        "# Convert predictions back to text\n",
        "predicted_texts = []\n",
        "for prediction in predictions:\n",
        "    predicted_text = ' '.join([tokenizer.index_word.get(idx, '') for idx in np.argmax(prediction, axis=1)])\n",
        "    predicted_texts.append(predicted_text)\n",
        "\n",
        "# Convert ground truth data back to text\n",
        "ground_truth_texts = test_target_texts\n",
        "\n",
        "# Calculate BLEU score\n",
        "references = [[text.split()] for text in ground_truth_texts]\n",
        "candidates = [text.split() for text in predicted_texts]\n",
        "bleu_score = corpus_bleu(references, candidates)\n",
        "print(f\"BLEU Score: {bleu_score * 100:.2f}%\")\n",
        "\n",
        "# Calculate BLEU score with smoothing\n",
        "smoothing_function = SmoothingFunction().method4\n",
        "bleu_score_smoothed = corpus_bleu(references, candidates, smoothing_function=smoothing_function)\n",
        "print(f\"BLEU Score with Smoothing: {bleu_score_smoothed * 100:.2f}%\")\n",
        "\n",
        "# Calculate ROUGE score\n",
        "reference_string = ' '.join(ground_truth_texts)\n",
        "hypothesis_string = ' '.join(predicted_texts)\n",
        "\n",
        "rouge_n_score = sentence_bleu(reference_string, hypothesis_string, weights=(1, 0, 0, 0))\n",
        "rouge_l_score = sentence_bleu(reference_string, hypothesis_string, weights=(0, 1, 0, 0))\n",
        "\n",
        "print(f\"ROUGE-N Score: {rouge_n_score * 100:.2f}%\")\n",
        "print(f\"ROUGE-L Score: {rouge_l_score * 100:.2f}%\")\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vwS89xIhgP_2"
      },
      "source": [
        "Evaluation on a random sentence using Beam Search Prediction"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "caZQAVi4gP_2",
        "outputId": "1c469b6e-2feb-4075-e40e-59e26ac23bc0"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "15/15 [==============================] - 3s 174ms/step\n",
            "Original Sentence: He is released back into the wild, now with more adult feathers, and a tracking device strapped to his back, and leads the humans home to his native colony.\n",
            "Actual Simplified Sentence: He is put back into the wild, now with more adult feathers, and a tracking device strapped to his back.  He leads the humans home to his native colony.\n",
            "Predicted Simplified Sentence: Is put back into the wild now with more adult feathers and a tracking device strapped to his back he leads the humans home to the native colony  of of of of of of of of of of of of of of of of.\n",
            "BLEU score: 0.15682101300022366\n",
            "BLEU score with smoothing: 0.15682101300022366\n",
            "ROUGE-N Score: 0.15%\n",
            "ROUGE-L Score: 0.00%\n"
          ]
        }
      ],
      "source": [
        "# Example of using beam search during prediction\n",
        "\n",
        "def beam_search_decoder(predictions, top_k=3):\n",
        "    start = [tokenizer.word_index['<start>']]\n",
        "\n",
        "    # Get the top K candidates at each step\n",
        "    sequences = [[start, 0.0]]\n",
        "    for _ in range(max_sequence_length):\n",
        "        all_candidates = []\n",
        "        for i in range(len(sequences)):\n",
        "            seq, score = sequences[i]\n",
        "            if seq[-1] == tokenizer.word_index['<end>']:\n",
        "                all_candidates.append([seq, score])\n",
        "                continue\n",
        "            expanded = model.predict([np.array([seq[-1]]), np.array([seq])])\n",
        "            predictions = expanded[0, -1, :]\n",
        "            word_preds = np.argsort(predictions)[::-1][:top_k]\n",
        "            for w in word_preds:\n",
        "                next_seq, next_score = seq + [w], score - np.log(predictions[w])\n",
        "                all_candidates.append([next_seq, next_score])\n",
        "        sequences = sorted(all_candidates, key=lambda x: x[1])[:top_k]\n",
        "\n",
        "    return sequences\n",
        "\n",
        "# Generate predictions using the trained model\n",
        "predictions = model.predict([test_encoder_input_data, test_decoder_input_data])\n",
        "\n",
        "# Convert predictions back to text\n",
        "predicted_texts = []\n",
        "for prediction in predictions:\n",
        "    predicted_text = ' '.join([tokenizer.index_word.get(idx, '') for idx in np.argmax(prediction, axis=1)])\n",
        "    predicted_text = predicted_text.replace('<start>', '').replace('<end>', '').strip()\n",
        "    predicted_text = predicted_text.capitalize() + '.'  # Capitalize first letter and add a period\n",
        "    predicted_texts.append(predicted_text)\n",
        "\n",
        "# Convert ground truth data back to text\n",
        "ground_truth_texts = test_target_texts\n",
        "\n",
        "# Print the original, actual simplified, and predicted simplified sentences for a random example\n",
        "index = 449\n",
        "input_sentence = test_input_texts[index]\n",
        "actual_simplified_sentence = test_target_texts[index]\n",
        "predicted_sentence = predicted_texts[index]\n",
        "\n",
        "print(\"Original Sentence:\", input_sentence)\n",
        "print(\"Actual Simplified Sentence:\", actual_simplified_sentence)\n",
        "print(\"Predicted Simplified Sentence:\", predicted_sentence)\n",
        "\n",
        "# Calculate BLEU score\n",
        "references = [[text.split()] for text in ground_truth_texts]\n",
        "candidates = [text.split() for text in predicted_texts]\n",
        "bleu_score = corpus_bleu(references, candidates)\n",
        "\n",
        "# Print the BLEU score\n",
        "print(f\"BLEU score: {bleu_score}\")\n",
        "\n",
        "# Calculate BLEU score with smoothing\n",
        "smoothing_function = SmoothingFunction().method4\n",
        "bleu_score = corpus_bleu(references, candidates, smoothing_function=smoothing_function)\n",
        "\n",
        "# Print the BLEU score with smoothing\n",
        "print(f\"BLEU score with smoothing: {bleu_score}\")\n",
        "\n",
        "# Calculate ROUGE score\n",
        "reference_string = ' '.join(ground_truth_texts)\n",
        "hypothesis_string = ' '.join(predicted_texts)\n",
        "\n",
        "rouge_n_score = sentence_bleu(reference_string, hypothesis_string, weights=(1, 0, 0, 0))\n",
        "rouge_l_score = sentence_bleu(reference_string, hypothesis_string, weights=(0, 1, 0, 0))\n",
        "\n",
        "print(f\"ROUGE-N Score: {rouge_n_score * 100:.2f}%\")\n",
        "print(f\"ROUGE-L Score: {rouge_l_score * 100:.2f}%\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "tcy-t_gKgP_2",
        "outputId": "56c6da19-a3c1-481e-c4d8-af96ad386620"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "15/15 [==============================] - 3s 177ms/step\n",
            "Original Sentence: Furthermore, she appeared in several music videos, including \"It Girl\" by John Oates and \"Just Lose It\" by Eminem.\n",
            "Actual Simplified Sentence: She was in music videos, including \"It Girl\" by John Oates and \"Just Lose It\" by Eminem.\n",
            "Predicted Simplified Sentence: Was popular several popular such popular was she she r and she he start was not have of of of of of of of of of of of of of of of of of of of of of of of of of of of of.\n",
            "BLEU score: 0.15682101300022366\n",
            "BLEU score with smoothing: 0.15682101300022366\n"
          ]
        },
        {
          "name": "stderr",
          "output_type": "stream",
          "text": [
            "/home/student/anaconda3/lib/python3.10/site-packages/nltk/translate/bleu_score.py:552: UserWarning: \n",
            "The hypothesis contains 0 counts of 2-gram overlaps.\n",
            "Therefore the BLEU score evaluates to 0, independently of\n",
            "how many N-gram overlaps of lower order it contains.\n",
            "Consider using lower n-gram order or use SmoothingFunction()\n",
            "  warnings.warn(_msg)\n",
            "/home/student/anaconda3/lib/python3.10/site-packages/nltk/translate/bleu_score.py:552: UserWarning: \n",
            "The hypothesis contains 0 counts of 3-gram overlaps.\n",
            "Therefore the BLEU score evaluates to 0, independently of\n",
            "how many N-gram overlaps of lower order it contains.\n",
            "Consider using lower n-gram order or use SmoothingFunction()\n",
            "  warnings.warn(_msg)\n",
            "/home/student/anaconda3/lib/python3.10/site-packages/nltk/translate/bleu_score.py:552: UserWarning: \n",
            "The hypothesis contains 0 counts of 4-gram overlaps.\n",
            "Therefore the BLEU score evaluates to 0, independently of\n",
            "how many N-gram overlaps of lower order it contains.\n",
            "Consider using lower n-gram order or use SmoothingFunction()\n",
            "  warnings.warn(_msg)\n"
          ]
        },
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "ROUGE-N Score: 0.15%\n",
            "ROUGE-L Score: 0.00%\n"
          ]
        }
      ],
      "source": [
        "# Example of using beam search during prediction\n",
        "\n",
        "def beam_search_decoder(predictions, top_k=3):\n",
        "    start = [tokenizer.word_index['<start>']]\n",
        "\n",
        "    # Get the top K candidates at each step\n",
        "    sequences = [[start, 0.0]]\n",
        "    for _ in range(max_sequence_length):\n",
        "        all_candidates = []\n",
        "        for i in range(len(sequences)):\n",
        "            seq, score = sequences[i]\n",
        "            if seq[-1] == tokenizer.word_index['<end>']:\n",
        "                all_candidates.append([seq, score])\n",
        "                continue\n",
        "            expanded = model.predict([np.array([seq[-1]]), np.array([seq])])\n",
        "            predictions = expanded[0, -1, :]\n",
        "            word_preds = np.argsort(predictions)[::-1][:top_k]\n",
        "            for w in word_preds:\n",
        "                next_seq, next_score = seq + [w], score - np.log(predictions[w])\n",
        "                all_candidates.append([next_seq, next_score])\n",
        "        sequences = sorted(all_candidates, key=lambda x: x[1])[:top_k]\n",
        "\n",
        "    return sequences\n",
        "\n",
        "# Generate predictions using the trained model\n",
        "predictions = model.predict([test_encoder_input_data, test_decoder_input_data])\n",
        "\n",
        "# Convert predictions back to text\n",
        "predicted_texts = []\n",
        "for prediction in predictions:\n",
        "    predicted_text = ' '.join([tokenizer.index_word.get(idx, '') for idx in np.argmax(prediction, axis=1)])\n",
        "    predicted_text = predicted_text.replace('<start>', '').replace('<end>', '').strip()\n",
        "    predicted_text = predicted_text.capitalize() + '.'  # Capitalize first letter and add a period\n",
        "    predicted_texts.append(predicted_text)\n",
        "\n",
        "# Convert ground truth data back to text\n",
        "ground_truth_texts = test_target_texts\n",
        "\n",
        "# Print the original, actual simplified, and predicted simplified sentences for a random example\n",
        "#index = np.random.randint(0, len(test_input_texts))\n",
        "index = 298\n",
        "input_sentence = test_input_texts[index]\n",
        "actual_simplified_sentence = test_target_texts[index]\n",
        "predicted_sentence = predicted_texts[index]\n",
        "\n",
        "print(\"Original Sentence:\", input_sentence)\n",
        "print(\"Actual Simplified Sentence:\", actual_simplified_sentence)\n",
        "print(\"Predicted Simplified Sentence:\", predicted_sentence)\n",
        "\n",
        "# Calculate BLEU score\n",
        "references = [[text.split()] for text in ground_truth_texts]\n",
        "candidates = [text.split() for text in predicted_texts]\n",
        "bleu_score = corpus_bleu(references, candidates)\n",
        "\n",
        "# Print the BLEU score\n",
        "print(f\"BLEU score: {bleu_score}\")\n",
        "\n",
        "# Calculate BLEU score with smoothing\n",
        "smoothing_function = SmoothingFunction().method4\n",
        "bleu_score = corpus_bleu(references, candidates, smoothing_function=smoothing_function)\n",
        "\n",
        "# Print the BLEU score with smoothing\n",
        "print(f\"BLEU score with smoothing: {bleu_score}\")\n",
        "\n",
        "# Calculate ROUGE score\n",
        "reference_string = ' '.join(ground_truth_texts)\n",
        "hypothesis_string = ' '.join(predicted_texts)\n",
        "\n",
        "rouge_n_score = sentence_bleu(reference_string, hypothesis_string, weights=(1, 0, 0, 0))\n",
        "rouge_l_score = sentence_bleu(reference_string, hypothesis_string, weights=(0, 1, 0, 0))\n",
        "\n",
        "print(f\"ROUGE-N Score: {rouge_n_score * 100:.2f}%\")\n",
        "print(f\"ROUGE-L Score: {rouge_l_score * 100:.2f}%\")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fuGXFHElgP_2"
      },
      "source": [
        "Evaluation of a random sentence"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "O0VOSzoEgP_2",
        "outputId": "a5a3af56-368a-4e9a-b6c9-0ea522d75808"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "1/1 [==============================] - 0s 86ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 97ms/step\n",
            "1/1 [==============================] - 0s 126ms/step\n",
            "1/1 [==============================] - 0s 152ms/step\n",
            "1/1 [==============================] - 0s 162ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 150ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 241ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 77ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 75ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 103ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 107ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 152ms/step\n",
            "1/1 [==============================] - 0s 167ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 88ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 126ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 107ms/step\n",
            "1/1 [==============================] - 0s 67ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 90ms/step\n",
            "1/1 [==============================] - 0s 78ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 145ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 111ms/step\n",
            "1/1 [==============================] - 0s 83ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 75ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 65ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 93ms/step\n",
            "1/1 [==============================] - 0s 103ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 62ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 91ms/step\n",
            "1/1 [==============================] - 0s 69ms/step\n",
            "1/1 [==============================] - 0s 65ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 100ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 132ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 56ms/step\n",
            "1/1 [==============================] - 0s 86ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 93ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 84ms/step\n",
            "1/1 [==============================] - 0s 96ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 91ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 73ms/step\n",
            "1/1 [==============================] - 0s 64ms/step\n",
            "1/1 [==============================] - 0s 100ms/step\n",
            "1/1 [==============================] - 0s 87ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 96ms/step\n",
            "1/1 [==============================] - 0s 97ms/step\n",
            "1/1 [==============================] - 0s 94ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 126ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 92ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 87ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 107ms/step\n",
            "1/1 [==============================] - 0s 66ms/step\n",
            "Original Sentence: In 1987 Wexler was inducted into the Rock and Roll Hall of Fame..\n",
            "Simplified Sentence (Ground Truth): In 1987 Wexler was recognized into the Rock and Roll Hall of Fame.\n",
            "Predicted Simplified Sentence: <oov> was put forth into the blues hall of fame hall and racehorse.\n",
            "BLEU Score: 0.00%\n",
            "BLEU Score with smoothing: 0.00%\n",
            "ROUGE-N Score: 38.46%\n",
            "ROUGE-L Score: 8.33%\n"
          ]
        }
      ],
      "source": [
        "import numpy as np\n",
        "import tensorflow as tf\n",
        "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
        "from nltk.translate.bleu_score import corpus_bleu, SmoothingFunction, sentence_bleu\n",
        "from nltk.translate.meteor_score import single_meteor_score\n",
        "\n",
        "# Manually set the index for '<START>' based on your training configuration\n",
        "start_token_index = 1  # Adjust this based on your training data\n",
        "oov_token_index = tokenizer.word_index.get('<OOV>', None) or len(tokenizer.word_index) + 1  # Get the index or assign a new one\n",
        "\n",
        "# Encode the input sentence\n",
        "test_input_sequence = tokenizer.texts_to_sequences([\"In 1987 Wexler was inducted into the Rock and Roll Hall of Fame.\"])\n",
        "encoder_test_input_data = pad_sequences(test_input_sequence, maxlen=max_sequence_length, padding='post')\n",
        "encoder_test_input_data = tf.convert_to_tensor(encoder_test_input_data, dtype=tf.float32)\n",
        "\n",
        "# Generate predictions using the trained model\n",
        "decoder_test_input_data = np.zeros((1, max_sequence_length))\n",
        "decoder_test_input_data[:, 0] = start_token_index  # Manually set the start token index\n",
        "for i in range(1, max_sequence_length):\n",
        "    predictions = model.predict([encoder_test_input_data, decoder_test_input_data])\n",
        "    predicted_token = np.argmax(predictions[0, i-1, :])\n",
        "    decoder_test_input_data[:, i] = predicted_token\n",
        "    if predicted_token == 0:\n",
        "        continue  # Skip the zero index\n",
        "    if tokenizer.index_word[predicted_token] == '<END>':\n",
        "        break\n",
        "\n",
        "# Convert predicted sequence to text, replacing OOV tokens\n",
        "predicted_text = [tokenizer.index_word.get(int(idx), '<OOV>') for idx in decoder_test_input_data[0] if idx != 0]\n",
        "predicted_text[0] = predicted_text[0].capitalize()\n",
        "predicted_text = \" \".join(predicted_text) + \".\"\n",
        "\n",
        "# Print the results\n",
        "original_sentence = \"In 1987 Wexler was inducted into the Rock and Roll Hall of Fame..\"\n",
        "simplified_sentence = \"In 1987 Wexler was recognized into the Rock and Roll Hall of Fame.\"\n",
        "print(\"Original Sentence:\", original_sentence)\n",
        "print(\"Simplified Sentence (Ground Truth):\", simplified_sentence)\n",
        "print(\"Predicted Simplified Sentence:\", predicted_text)\n",
        "\n",
        "# Calculate BLEU score\n",
        "reference = [simplified_sentence.split()]\n",
        "hypothesis = predicted_text.split()\n",
        "bleu_score = corpus_bleu(reference, [hypothesis])\n",
        "print(f\"BLEU Score: {bleu_score * 100:.2f}%\")\n",
        "\n",
        "# Calculate BLEU score with smoothing\n",
        "smoothed_bleu_score = corpus_bleu(reference, [hypothesis], smoothing_function=SmoothingFunction().method1)\n",
        "print(f\"BLEU Score with smoothing: {smoothed_bleu_score * 100:.2f}%\")\n",
        "\n",
        "# Calculate Rouge score\n",
        "rouge_n_score = sentence_bleu(reference, hypothesis, weights=(1, 0, 0, 0))\n",
        "rouge_l_score = sentence_bleu(reference, hypothesis, weights=(0, 1, 0, 0))\n",
        "\n",
        "print(f\"ROUGE-N Score: {rouge_n_score * 100:.2f}%\")\n",
        "print(f\"ROUGE-L Score: {rouge_l_score * 100:.2f}%\")\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UVwkEei5gP_2"
      },
      "source": [
        "Evaluation of a random sentence"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "6E6kWQLYgP_3",
        "outputId": "065e4be6-bf6d-4980-de4c-0cba82bd34b1"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "15/15 [==============================] - 2s 157ms/step\n",
            "Original Sentence: Offenbach's numerous operettas, such as Orpheus in the Underworld, and La belle Hélène, were extremely popular in both France and the English-speaking world during the 1850s and 1860s.\n",
            "Actual Simplified Sentence: Offenbach composed numerous operettas, such as Orpheus in the Underworld and La belle Helene.  His operettas were very popular in France and English-speaking countries in the 1850s and 1860s.\n",
            "Predicted Simplified Sentence: gave published churros in the 122 as san garden king paris english guard  current were many most by the  coptic  the  the english                  \n",
            "BLEU score: 0.16072221447412147\n",
            "BLEU score with smoothing: 0.16072221447412147\n",
            "ROUGE-N Score: 23.08%\n",
            "ROUGE-L Score: 0.00%\n"
          ]
        }
      ],
      "source": [
        "# Perform the necessary data preprocessing steps for the test data\n",
        "# Assuming you have a tokenizer\n",
        "# Tokenize the test data\n",
        "test_encoder_sequences = tokenizer.texts_to_sequences(test_input_texts)\n",
        "test_decoder_sequences = tokenizer.texts_to_sequences(test_target_texts)\n",
        "\n",
        "# Pad the sequences\n",
        "max_seq_length = max(max(len(seq) for seq in test_encoder_sequences), max(len(seq) for seq in test_decoder_sequences))\n",
        "test_encoder_input_data = pad_sequences(test_encoder_sequences, maxlen=max_seq_length, padding='post')\n",
        "test_decoder_input_data = pad_sequences(test_decoder_sequences, maxlen=max_seq_length, padding='post')\n",
        "\n",
        "# Use the preprocessed test data as the test encoder and decoder input data for evaluation\n",
        "test_encoder_input_data = test_encoder_input_data  # Adjust this based on your preprocessing\n",
        "test_decoder_input_data = test_decoder_input_data  # Adjust this based on your preprocessing\n",
        "\n",
        "\n",
        "# Evaluate the model using the test data\n",
        "test_encoder_input_data = tf.convert_to_tensor(test_encoder_input_data, dtype=tf.float32)\n",
        "test_decoder_input_data = tf.convert_to_tensor(test_decoder_input_data, dtype=tf.float32)\n",
        "\n",
        "# Generate predictions using the trained model\n",
        "predictions = model.predict([test_encoder_input_data, test_decoder_input_data])\n",
        "\n",
        "# Convert predictions back to text\n",
        "predicted_texts = []\n",
        "for prediction in predictions:\n",
        "    predicted_text = ' '.join([tokenizer.index_word.get(idx, '') for idx in np.argmax(prediction, axis=1)])\n",
        "    predicted_texts.append(predicted_text)\n",
        "\n",
        "# Convert ground truth data back to text\n",
        "ground_truth_texts = test_target_texts\n",
        "\n",
        "# Print the original, actual simplified, and predicted simplified sentences for a random example\n",
        "index = np.random.randint(0, len(test_input_texts))\n",
        "input_sentence = test_input_texts[index]\n",
        "actual_simplified_sentence = test_target_texts[index]\n",
        "predicted_sentence = predicted_texts[index]\n",
        "\n",
        "print(\"Original Sentence:\", input_sentence)\n",
        "print(\"Actual Simplified Sentence:\", actual_simplified_sentence)\n",
        "print(\"Predicted Simplified Sentence:\", predicted_sentence)\n",
        "\n",
        "\n",
        "# Calculate BLEU score\n",
        "references = [[text.split()] for text in ground_truth_texts]\n",
        "candidates = [text.split() for text in predicted_texts]\n",
        "bleu_score = corpus_bleu(references, candidates)\n",
        "\n",
        "# Print the BLEU score\n",
        "print(f\"BLEU score: {bleu_score}\")\n",
        "\n",
        "from nltk.translate.bleu_score import SmoothingFunction\n",
        "\n",
        "# Calculate BLEU score with smoothing\n",
        "smoothing_function = SmoothingFunction().method4\n",
        "bleu_score = corpus_bleu(references, candidates, smoothing_function=smoothing_function)\n",
        "\n",
        "# Print the BLEU score\n",
        "print(f\"BLEU score with smoothing: {bleu_score}\")\n",
        "\n",
        "# Calculate Rouge score\n",
        "rouge_n_score = sentence_bleu(reference, hypothesis, weights=(1, 0, 0, 0))\n",
        "rouge_l_score = sentence_bleu(reference, hypothesis, weights=(0, 1, 0, 0))\n",
        "\n",
        "print(f\"ROUGE-N Score: {rouge_n_score * 100:.2f}%\")\n",
        "print(f\"ROUGE-L Score: {rouge_l_score * 100:.2f}%\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Jctk1i7igP_3",
        "outputId": "25575995-2093-459c-d5c1-381c16e94386"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "1/1 [==============================] - 0s 97ms/step\n",
            "1/1 [==============================] - 0s 103ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 97ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 84ms/step\n",
            "1/1 [==============================] - 0s 65ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 206ms/step\n",
            "1/1 [==============================] - 0s 142ms/step\n",
            "1/1 [==============================] - 0s 126ms/step\n",
            "1/1 [==============================] - 0s 126ms/step\n",
            "1/1 [==============================] - 0s 153ms/step\n",
            "1/1 [==============================] - 0s 188ms/step\n",
            "1/1 [==============================] - 0s 145ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 100ms/step\n",
            "1/1 [==============================] - 0s 153ms/step\n",
            "1/1 [==============================] - 0s 190ms/step\n",
            "1/1 [==============================] - 0s 222ms/step\n",
            "1/1 [==============================] - 0s 240ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 212ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 92ms/step\n",
            "1/1 [==============================] - 0s 98ms/step\n",
            "1/1 [==============================] - 0s 106ms/step\n",
            "1/1 [==============================] - 0s 154ms/step\n",
            "1/1 [==============================] - 0s 95ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 91ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 169ms/step\n",
            "1/1 [==============================] - 0s 93ms/step\n",
            "1/1 [==============================] - 0s 178ms/step\n",
            "1/1 [==============================] - 0s 100ms/step\n",
            "1/1 [==============================] - 0s 91ms/step\n",
            "1/1 [==============================] - 0s 95ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 88ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 94ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 111ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 83ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 95ms/step\n",
            "1/1 [==============================] - 0s 81ms/step\n",
            "1/1 [==============================] - 0s 93ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 76ms/step\n",
            "1/1 [==============================] - 0s 61ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 95ms/step\n",
            "1/1 [==============================] - 0s 88ms/step\n",
            "1/1 [==============================] - 0s 92ms/step\n",
            "1/1 [==============================] - 0s 81ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 88ms/step\n",
            "1/1 [==============================] - 0s 158ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 237ms/step\n",
            "1/1 [==============================] - 0s 88ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 94ms/step\n",
            "1/1 [==============================] - 0s 96ms/step\n",
            "1/1 [==============================] - 0s 85ms/step\n",
            "1/1 [==============================] - 0s 132ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 107ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 87ms/step\n",
            "1/1 [==============================] - 0s 78ms/step\n",
            "1/1 [==============================] - 0s 68ms/step\n",
            "1/1 [==============================] - 0s 161ms/step\n",
            "1/1 [==============================] - 0s 100ms/step\n",
            "1/1 [==============================] - 0s 103ms/step\n",
            "1/1 [==============================] - 0s 107ms/step\n",
            "1/1 [==============================] - 0s 92ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 186ms/step\n",
            "1/1 [==============================] - 0s 94ms/step\n",
            "1/1 [==============================] - 0s 66ms/step\n",
            "1/1 [==============================] - 0s 145ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 97ms/step\n",
            "1/1 [==============================] - 0s 70ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 97ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 88ms/step\n",
            "1/1 [==============================] - 0s 93ms/step\n",
            "1/1 [==============================] - 0s 64ms/step\n",
            "1/1 [==============================] - 0s 71ms/step\n",
            "1/1 [==============================] - 0s 90ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 79ms/step\n",
            "1/1 [==============================] - 0s 90ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 164ms/step\n",
            "1/1 [==============================] - 0s 90ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "Original Sentence: Due to some delays, her release date, with remissions, is currently 12 April 2024.\n",
            "Simplified Sentence (Ground Truth): Her delayed release date is 12 April 2024.\n",
            "Predicted Simplified Sentence: <OOV> passes designs have 12 april 2024 with april 22\n",
            "BLEU Score: 0.00%\n",
            "BLEU Score with smoothing: 2.11%\n",
            "ROUGE-N Score: 10.00%\n",
            "ROUGE-L Score: 0.00%\n"
          ]
        }
      ],
      "source": [
        "import tensorflow as tf\n",
        "from nltk.translate.bleu_score import corpus_bleu, sentence_bleu, SmoothingFunction\n",
        "import pickle\n",
        "\n",
        "# Load the tokenizer used during training\n",
        "with open('tokenizer.pickle', 'rb') as handle:\n",
        "    tokenizer = pickle.load(handle)\n",
        "\n",
        "# Get the index or assign a new one for the '<OOV>' token\n",
        "oov_token_index = tokenizer.word_index.get('<OOV>', None) or len(tokenizer.word_index) + 1\n",
        "\n",
        "# Manually set the index for '<START>' based on your training configuration\n",
        "start_token_index = 1  # Adjust this based on your training data\n",
        "\n",
        "# Encode the input sentence\n",
        "test_input_sequence = tokenizer.texts_to_sequences([\"Due to some delays, her release date, with remissions, is currently 12 April 2024.\"])\n",
        "encoder_test_input_data = pad_sequences(test_input_sequence, maxlen=max_sequence_length, padding='post')\n",
        "encoder_test_input_data = tf.convert_to_tensor(encoder_test_input_data, dtype=tf.float32)\n",
        "\n",
        "# Generate predictions using the trained model\n",
        "decoder_test_input_data = np.zeros((1, max_sequence_length))\n",
        "decoder_test_input_data[:, 0] = start_token_index  # Manually set the start token index\n",
        "for i in range(1, max_sequence_length):\n",
        "    predictions = model.predict([encoder_test_input_data, decoder_test_input_data])\n",
        "    predicted_token = np.argmax(predictions[0, i-1, :])\n",
        "    decoder_test_input_data[:, i] = predicted_token\n",
        "    if predicted_token == 0:\n",
        "        continue  # Skip the zero index\n",
        "    if tokenizer.index_word[predicted_token] == '<END>':\n",
        "        break\n",
        "\n",
        "# Convert predicted sequence to text, replacing OOV tokens\n",
        "predicted_text = \" \".join([tokenizer.index_word.get(int(idx), '<OOV>') for idx in decoder_test_input_data[0] if idx != 0])\n",
        "\n",
        "# Print the results\n",
        "original_sentence = \"Due to some delays, her release date, with remissions, is currently 12 April 2024.\"\n",
        "simplified_sentence = \"Her delayed release date is 12 April 2024.\"\n",
        "print(\"Original Sentence:\", original_sentence)\n",
        "print(\"Simplified Sentence (Ground Truth):\", simplified_sentence)\n",
        "print(\"Predicted Simplified Sentence:\", predicted_text)\n",
        "\n",
        "# Calculate BLEU score\n",
        "reference = [simplified_sentence.split()]\n",
        "hypothesis = predicted_text.split()\n",
        "bleu_score = sentence_bleu(reference, hypothesis)\n",
        "print(f\"BLEU Score: {bleu_score * 100:.2f}%\")\n",
        "\n",
        "# Calculate BLEU score with smoothing\n",
        "smoothed_bleu_score = sentence_bleu(reference, hypothesis, smoothing_function=SmoothingFunction().method1)\n",
        "print(f\"BLEU Score with smoothing: {smoothed_bleu_score * 100:.2f}%\")\n",
        "\n",
        "# Calculate Rouge score\n",
        "rouge_n_score = sentence_bleu(reference, hypothesis, weights=(1, 0, 0, 0))\n",
        "rouge_l_score = sentence_bleu(reference, hypothesis, weights=(0, 1, 0, 0))\n",
        "\n",
        "print(f\"ROUGE-N Score: {rouge_n_score * 100:.2f}%\")\n",
        "print(f\"ROUGE-L Score: {rouge_l_score * 100:.2f}%\")"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1yIFSWyugP_6"
      },
      "source": [
        "Infering from unknown sentences of Research paper"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "LZXS4rIdgP_6",
        "outputId": "62ca44b6-76b0-45c8-e29e-43c079026caa"
      },
      "outputs": [
        {
          "name": "stdout",
          "output_type": "stream",
          "text": [
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 201ms/step\n",
            "1/1 [==============================] - 0s 146ms/step\n",
            "1/1 [==============================] - 0s 175ms/step\n",
            "1/1 [==============================] - 0s 107ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 155ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 86ms/step\n",
            "1/1 [==============================] - 0s 111ms/step\n",
            "1/1 [==============================] - 0s 92ms/step\n",
            "1/1 [==============================] - 0s 87ms/step\n",
            "1/1 [==============================] - 0s 89ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 168ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 87ms/step\n",
            "1/1 [==============================] - 0s 73ms/step\n",
            "1/1 [==============================] - 0s 96ms/step\n",
            "1/1 [==============================] - 0s 72ms/step\n",
            "1/1 [==============================] - 0s 75ms/step\n",
            "1/1 [==============================] - 0s 81ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 96ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 142ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 85ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 78ms/step\n",
            "1/1 [==============================] - 0s 79ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 81ms/step\n",
            "1/1 [==============================] - 0s 75ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 80ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 188ms/step\n",
            "1/1 [==============================] - 0s 71ms/step\n",
            "1/1 [==============================] - 0s 171ms/step\n",
            "1/1 [==============================] - 0s 132ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 184ms/step\n",
            "1/1 [==============================] - 0s 107ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 155ms/step\n",
            "1/1 [==============================] - 0s 174ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 201ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 191ms/step\n",
            "1/1 [==============================] - 0s 80ms/step\n",
            "1/1 [==============================] - 0s 169ms/step\n",
            "1/1 [==============================] - 0s 176ms/step\n",
            "1/1 [==============================] - 0s 163ms/step\n",
            "1/1 [==============================] - 0s 246ms/step\n",
            "1/1 [==============================] - 0s 150ms/step\n",
            "1/1 [==============================] - 0s 173ms/step\n",
            "1/1 [==============================] - 0s 160ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 168ms/step\n",
            "1/1 [==============================] - 0s 159ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 153ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 96ms/step\n",
            "1/1 [==============================] - 0s 156ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 218ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 168ms/step\n",
            "1/1 [==============================] - 0s 92ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 126ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 208ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 161ms/step\n",
            "1/1 [==============================] - 0s 173ms/step\n",
            "1/1 [==============================] - 0s 219ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 157ms/step\n",
            "1/1 [==============================] - 0s 80ms/step\n",
            "1/1 [==============================] - 0s 111ms/step\n",
            "1/1 [==============================] - 0s 96ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 74ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 71ms/step\n",
            "1/1 [==============================] - 0s 86ms/step\n",
            "1/1 [==============================] - 0s 77ms/step\n",
            "1/1 [==============================] - 0s 79ms/step\n",
            "1/1 [==============================] - 0s 176ms/step\n",
            "1/1 [==============================] - 0s 111ms/step\n",
            "1/1 [==============================] - 0s 81ms/step\n",
            "1/1 [==============================] - 0s 203ms/step\n",
            "1/1 [==============================] - 0s 159ms/step\n",
            "1/1 [==============================] - 0s 166ms/step\n",
            "1/1 [==============================] - 0s 259ms/step\n",
            "1/1 [==============================] - 0s 226ms/step\n",
            "1/1 [==============================] - 0s 184ms/step\n",
            "1/1 [==============================] - 0s 158ms/step\n",
            "1/1 [==============================] - 0s 211ms/step\n",
            "1/1 [==============================] - 0s 163ms/step\n",
            "1/1 [==============================] - 0s 170ms/step\n",
            "1/1 [==============================] - 0s 169ms/step\n",
            "1/1 [==============================] - 0s 391ms/step\n",
            "1/1 [==============================] - 0s 154ms/step\n",
            "1/1 [==============================] - 0s 156ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 193ms/step\n",
            "1/1 [==============================] - 0s 176ms/step\n",
            "1/1 [==============================] - 0s 193ms/step\n",
            "1/1 [==============================] - 0s 172ms/step\n",
            "1/1 [==============================] - 0s 145ms/step\n",
            "1/1 [==============================] - 0s 178ms/step\n",
            "1/1 [==============================] - 0s 195ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 155ms/step\n",
            "1/1 [==============================] - 0s 189ms/step\n",
            "1/1 [==============================] - 0s 160ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 150ms/step\n",
            "1/1 [==============================] - 0s 152ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 159ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 165ms/step\n",
            "1/1 [==============================] - 0s 153ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 126ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 231ms/step\n",
            "1/1 [==============================] - 0s 205ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 87ms/step\n",
            "1/1 [==============================] - 0s 204ms/step\n",
            "1/1 [==============================] - 0s 163ms/step\n",
            "1/1 [==============================] - 0s 198ms/step\n",
            "1/1 [==============================] - 0s 166ms/step\n",
            "1/1 [==============================] - 0s 156ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 154ms/step\n",
            "1/1 [==============================] - 0s 98ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 100ms/step\n",
            "1/1 [==============================] - 0s 111ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 135ms/step\n",
            "1/1 [==============================] - 0s 160ms/step\n",
            "1/1 [==============================] - 0s 180ms/step\n",
            "1/1 [==============================] - 0s 167ms/step\n",
            "1/1 [==============================] - 0s 146ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 135ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 97ms/step\n",
            "1/1 [==============================] - 0s 84ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 106ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 160ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 90ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 182ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 103ms/step\n",
            "1/1 [==============================] - 0s 64ms/step\n",
            "1/1 [==============================] - 0s 70ms/step\n",
            "1/1 [==============================] - 0s 87ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 163ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 149ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 191ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 73ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 103ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 82ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 257ms/step\n",
            "1/1 [==============================] - 0s 183ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 229ms/step\n",
            "1/1 [==============================] - 0s 231ms/step\n",
            "1/1 [==============================] - 0s 366ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 301ms/step\n",
            "1/1 [==============================] - 0s 168ms/step\n",
            "1/1 [==============================] - 0s 181ms/step\n",
            "1/1 [==============================] - 0s 195ms/step\n",
            "1/1 [==============================] - 0s 166ms/step\n",
            "1/1 [==============================] - 0s 146ms/step\n",
            "1/1 [==============================] - 0s 254ms/step\n",
            "1/1 [==============================] - 0s 313ms/step\n",
            "1/1 [==============================] - 0s 150ms/step\n",
            "1/1 [==============================] - 0s 93ms/step\n",
            "1/1 [==============================] - 0s 176ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 184ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 367ms/step\n",
            "1/1 [==============================] - 0s 184ms/step\n",
            "1/1 [==============================] - 0s 211ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 227ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 436ms/step\n",
            "1/1 [==============================] - 0s 226ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 154ms/step\n",
            "1/1 [==============================] - 0s 300ms/step\n",
            "1/1 [==============================] - 0s 194ms/step\n",
            "1/1 [==============================] - 0s 165ms/step\n",
            "1/1 [==============================] - 0s 428ms/step\n",
            "1/1 [==============================] - 0s 156ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 106ms/step\n",
            "1/1 [==============================] - 0s 160ms/step\n",
            "1/1 [==============================] - 0s 249ms/step\n",
            "1/1 [==============================] - 0s 176ms/step\n",
            "1/1 [==============================] - 0s 292ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 211ms/step\n",
            "1/1 [==============================] - 0s 216ms/step\n",
            "1/1 [==============================] - 0s 185ms/step\n",
            "1/1 [==============================] - 0s 107ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 159ms/step\n",
            "1/1 [==============================] - 0s 285ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 186ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 149ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 177ms/step\n",
            "1/1 [==============================] - 0s 182ms/step\n",
            "1/1 [==============================] - 0s 111ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 178ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 166ms/step\n",
            "1/1 [==============================] - 0s 160ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 165ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 146ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 100ms/step\n",
            "1/1 [==============================] - 0s 97ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 90ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 156ms/step\n",
            "1/1 [==============================] - 0s 165ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 142ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 178ms/step\n",
            "1/1 [==============================] - 0s 168ms/step\n",
            "1/1 [==============================] - 0s 158ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 161ms/step\n",
            "1/1 [==============================] - 0s 230ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 166ms/step\n",
            "1/1 [==============================] - 0s 157ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 158ms/step\n",
            "1/1 [==============================] - 0s 186ms/step\n",
            "1/1 [==============================] - 0s 173ms/step\n",
            "1/1 [==============================] - 0s 277ms/step\n",
            "1/1 [==============================] - 0s 405ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 214ms/step\n",
            "1/1 [==============================] - 0s 145ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 94ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 160ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 152ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 142ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 165ms/step\n",
            "1/1 [==============================] - 0s 194ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 98ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 158ms/step\n",
            "1/1 [==============================] - 0s 189ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 149ms/step\n",
            "1/1 [==============================] - 0s 153ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 179ms/step\n",
            "1/1 [==============================] - 0s 166ms/step\n",
            "1/1 [==============================] - 0s 167ms/step\n",
            "1/1 [==============================] - 0s 166ms/step\n",
            "1/1 [==============================] - 0s 156ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 180ms/step\n",
            "1/1 [==============================] - 0s 145ms/step\n",
            "1/1 [==============================] - 0s 173ms/step\n",
            "1/1 [==============================] - 0s 159ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 153ms/step\n",
            "1/1 [==============================] - 0s 142ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 135ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 149ms/step\n",
            "1/1 [==============================] - 0s 154ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 206ms/step\n",
            "1/1 [==============================] - 0s 162ms/step\n",
            "1/1 [==============================] - 0s 157ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 169ms/step\n",
            "1/1 [==============================] - 0s 172ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 242ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 107ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 146ms/step\n",
            "1/1 [==============================] - 0s 165ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 82ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 93ms/step\n",
            "1/1 [==============================] - 0s 103ms/step\n",
            "1/1 [==============================] - 0s 76ms/step\n",
            "1/1 [==============================] - 0s 185ms/step\n",
            "1/1 [==============================] - 0s 91ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 81ms/step\n",
            "1/1 [==============================] - 0s 146ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 205ms/step\n",
            "1/1 [==============================] - 0s 180ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 91ms/step\n",
            "1/1 [==============================] - 0s 175ms/step\n",
            "1/1 [==============================] - 0s 347ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 364ms/step\n",
            "1/1 [==============================] - 1s 605ms/step\n",
            "1/1 [==============================] - 0s 196ms/step\n",
            "1/1 [==============================] - 0s 232ms/step\n",
            "1/1 [==============================] - 0s 198ms/step\n",
            "1/1 [==============================] - 0s 249ms/step\n",
            "1/1 [==============================] - 0s 94ms/step\n",
            "1/1 [==============================] - 0s 198ms/step\n",
            "1/1 [==============================] - 0s 266ms/step\n",
            "1/1 [==============================] - 0s 180ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 343ms/step\n",
            "1/1 [==============================] - 0s 249ms/step\n",
            "1/1 [==============================] - 0s 182ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 206ms/step\n",
            "1/1 [==============================] - 0s 153ms/step\n",
            "1/1 [==============================] - 0s 176ms/step\n",
            "1/1 [==============================] - 0s 361ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 156ms/step\n",
            "1/1 [==============================] - 0s 297ms/step\n",
            "1/1 [==============================] - 0s 83ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 285ms/step\n",
            "1/1 [==============================] - 0s 111ms/step\n",
            "1/1 [==============================] - 0s 158ms/step\n",
            "1/1 [==============================] - 0s 163ms/step\n",
            "1/1 [==============================] - 0s 357ms/step\n",
            "1/1 [==============================] - 0s 271ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 178ms/step\n",
            "1/1 [==============================] - 0s 339ms/step\n",
            "1/1 [==============================] - 0s 152ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 174ms/step\n",
            "1/1 [==============================] - 0s 167ms/step\n",
            "1/1 [==============================] - 0s 165ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 149ms/step\n",
            "1/1 [==============================] - 0s 165ms/step\n",
            "1/1 [==============================] - 0s 167ms/step\n",
            "1/1 [==============================] - 0s 160ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 93ms/step\n",
            "1/1 [==============================] - 0s 272ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 81ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 135ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 1s 500ms/step\n",
            "1/1 [==============================] - 0s 132ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 80ms/step\n",
            "1/1 [==============================] - 0s 87ms/step\n",
            "1/1 [==============================] - 0s 95ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 86ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 97ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 318ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 132ms/step\n",
            "1/1 [==============================] - 0s 211ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 166ms/step\n",
            "1/1 [==============================] - 0s 178ms/step\n",
            "1/1 [==============================] - 0s 192ms/step\n",
            "1/1 [==============================] - 0s 310ms/step\n",
            "1/1 [==============================] - 0s 200ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 345ms/step\n",
            "1/1 [==============================] - 0s 283ms/step\n",
            "1/1 [==============================] - 0s 167ms/step\n",
            "1/1 [==============================] - 0s 154ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 208ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 215ms/step\n",
            "1/1 [==============================] - 0s 162ms/step\n",
            "1/1 [==============================] - 0s 252ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 223ms/step\n",
            "1/1 [==============================] - 0s 171ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 156ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 205ms/step\n",
            "1/1 [==============================] - 0s 452ms/step\n",
            "1/1 [==============================] - 0s 238ms/step\n",
            "1/1 [==============================] - 0s 241ms/step\n",
            "1/1 [==============================] - 0s 153ms/step\n",
            "1/1 [==============================] - 0s 248ms/step\n",
            "1/1 [==============================] - 0s 178ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 445ms/step\n",
            "1/1 [==============================] - 0s 158ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 152ms/step\n",
            "1/1 [==============================] - 0s 100ms/step\n",
            "1/1 [==============================] - 0s 90ms/step\n",
            "1/1 [==============================] - 1s 529ms/step\n",
            "1/1 [==============================] - 0s 157ms/step\n",
            "1/1 [==============================] - 0s 222ms/step\n",
            "1/1 [==============================] - 0s 212ms/step\n",
            "1/1 [==============================] - 0s 220ms/step\n",
            "1/1 [==============================] - 0s 165ms/step\n",
            "1/1 [==============================] - 0s 149ms/step\n",
            "1/1 [==============================] - 0s 183ms/step\n",
            "1/1 [==============================] - 0s 374ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 453ms/step\n",
            "1/1 [==============================] - 0s 218ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 162ms/step\n",
            "1/1 [==============================] - 0s 135ms/step\n",
            "1/1 [==============================] - 0s 177ms/step\n",
            "1/1 [==============================] - 0s 341ms/step\n",
            "1/1 [==============================] - 0s 132ms/step\n",
            "1/1 [==============================] - 0s 163ms/step\n",
            "1/1 [==============================] - 0s 150ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 135ms/step\n",
            "1/1 [==============================] - 0s 111ms/step\n",
            "1/1 [==============================] - 0s 230ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 175ms/step\n",
            "1/1 [==============================] - 0s 179ms/step\n",
            "1/1 [==============================] - 0s 254ms/step\n",
            "1/1 [==============================] - 0s 239ms/step\n",
            "1/1 [==============================] - 0s 224ms/step\n",
            "1/1 [==============================] - 0s 153ms/step\n",
            "1/1 [==============================] - 0s 327ms/step\n",
            "1/1 [==============================] - 0s 407ms/step\n",
            "1/1 [==============================] - 0s 311ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 161ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 100ms/step\n",
            "1/1 [==============================] - 0s 100ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 80ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 94ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 159ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 142ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 170ms/step\n",
            "1/1 [==============================] - 0s 159ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 154ms/step\n",
            "1/1 [==============================] - 0s 132ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 213ms/step\n",
            "1/1 [==============================] - 0s 160ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 168ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 152ms/step\n",
            "1/1 [==============================] - 0s 126ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 205ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 90ms/step\n",
            "1/1 [==============================] - 0s 162ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 96ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 97ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 153ms/step\n",
            "1/1 [==============================] - 0s 126ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 167ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 93ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 152ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 146ms/step\n",
            "1/1 [==============================] - 0s 175ms/step\n",
            "1/1 [==============================] - 0s 201ms/step\n",
            "1/1 [==============================] - 0s 399ms/step\n",
            "1/1 [==============================] - 0s 308ms/step\n",
            "1/1 [==============================] - 0s 237ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 203ms/step\n",
            "1/1 [==============================] - 0s 183ms/step\n",
            "1/1 [==============================] - 0s 159ms/step\n",
            "1/1 [==============================] - 0s 160ms/step\n",
            "1/1 [==============================] - 0s 180ms/step\n",
            "1/1 [==============================] - 0s 106ms/step\n",
            "1/1 [==============================] - 0s 155ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 162ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 154ms/step\n",
            "1/1 [==============================] - 0s 171ms/step\n",
            "1/1 [==============================] - 0s 170ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 157ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 154ms/step\n",
            "1/1 [==============================] - 0s 162ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 154ms/step\n",
            "1/1 [==============================] - 0s 155ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 165ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 149ms/step\n",
            "1/1 [==============================] - 0s 156ms/step\n",
            "1/1 [==============================] - 0s 87ms/step\n",
            "1/1 [==============================] - 0s 200ms/step\n",
            "1/1 [==============================] - 0s 64ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 103ms/step\n",
            "1/1 [==============================] - 0s 167ms/step\n",
            "1/1 [==============================] - 0s 167ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 244ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 149ms/step\n",
            "1/1 [==============================] - 0s 173ms/step\n",
            "1/1 [==============================] - 0s 193ms/step\n",
            "1/1 [==============================] - 0s 161ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 161ms/step\n",
            "1/1 [==============================] - 0s 159ms/step\n",
            "1/1 [==============================] - 0s 177ms/step\n",
            "1/1 [==============================] - 0s 164ms/step\n",
            "1/1 [==============================] - 0s 222ms/step\n",
            "1/1 [==============================] - 0s 158ms/step\n",
            "1/1 [==============================] - 0s 171ms/step\n",
            "1/1 [==============================] - 0s 174ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 241ms/step\n",
            "1/1 [==============================] - 0s 154ms/step\n",
            "1/1 [==============================] - 0s 166ms/step\n",
            "1/1 [==============================] - 0s 215ms/step\n",
            "1/1 [==============================] - 0s 265ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 165ms/step\n",
            "1/1 [==============================] - 0s 166ms/step\n",
            "1/1 [==============================] - 0s 175ms/step\n",
            "1/1 [==============================] - 0s 174ms/step\n",
            "1/1 [==============================] - 0s 146ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 152ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 156ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 159ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 187ms/step\n",
            "1/1 [==============================] - 0s 171ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 145ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 155ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 300ms/step\n",
            "1/1 [==============================] - 0s 396ms/step\n",
            "1/1 [==============================] - 0s 154ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 403ms/step\n",
            "1/1 [==============================] - 0s 213ms/step\n",
            "1/1 [==============================] - 0s 232ms/step\n",
            "1/1 [==============================] - 0s 291ms/step\n",
            "1/1 [==============================] - 0s 201ms/step\n",
            "1/1 [==============================] - 0s 226ms/step\n",
            "1/1 [==============================] - 0s 202ms/step\n",
            "1/1 [==============================] - 0s 292ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 280ms/step\n",
            "1/1 [==============================] - 0s 173ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 169ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 161ms/step\n",
            "1/1 [==============================] - 0s 163ms/step\n",
            "1/1 [==============================] - 0s 180ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 154ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 227ms/step\n",
            "1/1 [==============================] - 0s 126ms/step\n",
            "1/1 [==============================] - 0s 238ms/step\n",
            "1/1 [==============================] - 0s 153ms/step\n",
            "1/1 [==============================] - 0s 264ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 259ms/step\n",
            "1/1 [==============================] - 0s 193ms/step\n",
            "1/1 [==============================] - 0s 142ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 142ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 103ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 182ms/step\n",
            "1/1 [==============================] - 0s 159ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 142ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 69ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 157ms/step\n",
            "1/1 [==============================] - 0s 90ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 146ms/step\n",
            "1/1 [==============================] - 0s 197ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 165ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 107ms/step\n",
            "1/1 [==============================] - 0s 164ms/step\n",
            "1/1 [==============================] - 0s 162ms/step\n",
            "1/1 [==============================] - 0s 173ms/step\n",
            "1/1 [==============================] - 0s 162ms/step\n",
            "1/1 [==============================] - 0s 180ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 187ms/step\n",
            "1/1 [==============================] - 0s 159ms/step\n",
            "1/1 [==============================] - 0s 166ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 146ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 155ms/step\n",
            "1/1 [==============================] - 0s 163ms/step\n",
            "1/1 [==============================] - 0s 106ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 218ms/step\n",
            "1/1 [==============================] - 0s 96ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 75ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 200ms/step\n",
            "1/1 [==============================] - 0s 228ms/step\n",
            "1/1 [==============================] - 0s 188ms/step\n",
            "1/1 [==============================] - 0s 107ms/step\n",
            "1/1 [==============================] - 0s 156ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 126ms/step\n",
            "1/1 [==============================] - 0s 98ms/step\n",
            "1/1 [==============================] - 0s 98ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 97ms/step\n",
            "1/1 [==============================] - 0s 178ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 186ms/step\n",
            "1/1 [==============================] - 0s 171ms/step\n",
            "1/1 [==============================] - 0s 164ms/step\n",
            "1/1 [==============================] - 0s 175ms/step\n",
            "1/1 [==============================] - 0s 155ms/step\n",
            "1/1 [==============================] - 0s 110ms/step\n",
            "1/1 [==============================] - 0s 177ms/step\n",
            "1/1 [==============================] - 0s 157ms/step\n",
            "1/1 [==============================] - 0s 161ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 172ms/step\n",
            "1/1 [==============================] - 0s 93ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 126ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 152ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 160ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 161ms/step\n",
            "1/1 [==============================] - 0s 145ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 173ms/step\n",
            "1/1 [==============================] - 0s 239ms/step\n",
            "1/1 [==============================] - 0s 158ms/step\n",
            "1/1 [==============================] - 0s 149ms/step\n",
            "1/1 [==============================] - 0s 103ms/step\n",
            "1/1 [==============================] - 0s 82ms/step\n",
            "1/1 [==============================] - 0s 68ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 75ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 185ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 142ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 74ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 96ms/step\n",
            "1/1 [==============================] - 0s 88ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 135ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 153ms/step\n",
            "1/1 [==============================] - 0s 199ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 167ms/step\n",
            "1/1 [==============================] - 0s 89ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 145ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 172ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 129ms/step\n",
            "1/1 [==============================] - 0s 182ms/step\n",
            "1/1 [==============================] - 0s 186ms/step\n",
            "1/1 [==============================] - 0s 179ms/step\n",
            "1/1 [==============================] - 0s 150ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 85ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 195ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 88ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 189ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 166ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 126ms/step\n",
            "1/1 [==============================] - 0s 170ms/step\n",
            "1/1 [==============================] - 0s 83ms/step\n",
            "1/1 [==============================] - 0s 77ms/step\n",
            "1/1 [==============================] - 0s 126ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 157ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 168ms/step\n",
            "1/1 [==============================] - 0s 154ms/step\n",
            "1/1 [==============================] - 0s 157ms/step\n",
            "1/1 [==============================] - 0s 164ms/step\n",
            "1/1 [==============================] - 0s 88ms/step\n",
            "1/1 [==============================] - 0s 106ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 90ms/step\n",
            "1/1 [==============================] - 0s 68ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 107ms/step\n",
            "1/1 [==============================] - 0s 99ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 92ms/step\n",
            "1/1 [==============================] - 0s 111ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 85ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 208ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 97ms/step\n",
            "1/1 [==============================] - 0s 89ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 95ms/step\n",
            "1/1 [==============================] - 0s 90ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 107ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 106ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 106ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 128ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 166ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 152ms/step\n",
            "1/1 [==============================] - 0s 140ms/step\n",
            "1/1 [==============================] - 0s 94ms/step\n",
            "1/1 [==============================] - 0s 143ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 121ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 68ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 132ms/step\n",
            "1/1 [==============================] - 0s 163ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 89ms/step\n",
            "1/1 [==============================] - 0s 153ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 111ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 89ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 107ms/step\n",
            "1/1 [==============================] - 0s 98ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 91ms/step\n",
            "1/1 [==============================] - 0s 83ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 106ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 146ms/step\n",
            "1/1 [==============================] - 0s 152ms/step\n",
            "1/1 [==============================] - 0s 224ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 136ms/step\n",
            "1/1 [==============================] - 0s 228ms/step\n",
            "1/1 [==============================] - 0s 106ms/step\n",
            "1/1 [==============================] - 0s 97ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 147ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 131ms/step\n",
            "1/1 [==============================] - 0s 137ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 85ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 118ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 117ms/step\n",
            "1/1 [==============================] - 0s 69ms/step\n",
            "1/1 [==============================] - 0s 127ms/step\n",
            "1/1 [==============================] - 0s 134ms/step\n",
            "1/1 [==============================] - 0s 96ms/step\n",
            "1/1 [==============================] - 0s 273ms/step\n",
            "1/1 [==============================] - 0s 120ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 105ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 92ms/step\n",
            "1/1 [==============================] - 0s 141ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 181ms/step\n",
            "1/1 [==============================] - 0s 187ms/step\n",
            "1/1 [==============================] - 0s 102ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 93ms/step\n",
            "1/1 [==============================] - 0s 101ms/step\n",
            "1/1 [==============================] - 0s 139ms/step\n",
            "1/1 [==============================] - 0s 144ms/step\n",
            "1/1 [==============================] - 0s 130ms/step\n",
            "1/1 [==============================] - 0s 103ms/step\n",
            "1/1 [==============================] - 0s 100ms/step\n",
            "1/1 [==============================] - 0s 138ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 175ms/step\n",
            "1/1 [==============================] - 0s 76ms/step\n",
            "1/1 [==============================] - 0s 133ms/step\n",
            "1/1 [==============================] - 0s 123ms/step\n",
            "1/1 [==============================] - 0s 183ms/step\n",
            "1/1 [==============================] - 0s 87ms/step\n",
            "1/1 [==============================] - 0s 81ms/step\n",
            "1/1 [==============================] - 0s 112ms/step\n",
            "1/1 [==============================] - 0s 151ms/step\n",
            "1/1 [==============================] - 0s 83ms/step\n",
            "1/1 [==============================] - 0s 116ms/step\n",
            "1/1 [==============================] - 0s 145ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 179ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 122ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 104ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "1/1 [==============================] - 0s 97ms/step\n",
            "1/1 [==============================] - 0s 148ms/step\n",
            "1/1 [==============================] - 0s 222ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 155ms/step\n",
            "1/1 [==============================] - 0s 162ms/step\n",
            "1/1 [==============================] - 0s 119ms/step\n",
            "1/1 [==============================] - 0s 113ms/step\n",
            "1/1 [==============================] - 0s 115ms/step\n",
            "1/1 [==============================] - 0s 73ms/step\n",
            "1/1 [==============================] - 0s 84ms/step\n",
            "1/1 [==============================] - 0s 94ms/step\n",
            "1/1 [==============================] - 0s 125ms/step\n",
            "1/1 [==============================] - 0s 103ms/step\n",
            "1/1 [==============================] - 0s 124ms/step\n",
            "1/1 [==============================] - 0s 109ms/step\n",
            "1/1 [==============================] - 0s 83ms/step\n",
            "1/1 [==============================] - 0s 111ms/step\n",
            "1/1 [==============================] - 0s 114ms/step\n",
            "1/1 [==============================] - 0s 108ms/step\n",
            "Original Sentence: The shortest path problem based on the data structure has become one of the hot research topics in graph theory.\n",
            "Predicted Simplified Sentence: It Is Become A Single Culture For The Most Of The Most System And Best Of World.\n",
            "\n",
            "Original Sentence: As the basic theory of solving this problem, Dijkstra algorithm has been widely used in engineering calculations.\n",
            "Predicted Simplified Sentence: The Igneous Theory Method Has Usually Be Also For Reduced Instruction Has Been Needed.\n",
            "\n",
            "Original Sentence: Aiming at the shortcomings of traditional Dijkstra algorithm, this paper has proposed an optimization method which has mainly improved the nodes selection of the shortest path and data storage structure and organization.\n",
            "Predicted Simplified Sentence: A Lot Of Slavery That 1835 A Iso Concrete But It Was Considered An Mass Of Makes It Possibly To Be The First Distribution.\n",
            "\n",
            "Original Sentence: Through comparison and analysis, the improved algorithm has been obtained, which has reduced the storage space, improved the operational efficiency and has a better applicability in the shortest path calculation.\n",
            "Predicted Simplified Sentence: Before Its Exist Pressure Has Been Arrested Unit To Development Of Its Lines Oecussi Compatibility And An Development Of That System.\n",
            "\n",
            "Original Sentence: People's requirements for living and production efficiency have improved with the rapid development of computer science, in which the shortest path problem has become more prominent, and gradually becomes one of the hot topics of geographic information science, operations research, computer science and other disciplines.\n",
            "Predicted Simplified Sentence: Pipelining Has A Large Term That Has A Lot Of Debate System In Common And Codes Is A Most Of A Free Of A Free Of Rss Procedure A Development Of Procedure Top.\n",
            "\n",
            "Original Sentence: It is not only the key problem in network analysis but also the key issue in graph theory, and therefore it has been widely used in electronic navigation, network optimization, logistics, transportation and other areas in people's daily life.\n",
            "Predicted Simplified Sentence: That Is Not The Same In The Free Of Portions Of Which Has Been Used In Scratching Papyrus.\n",
            "\n",
            "Original Sentence: The shortest path is also the one between the nodes in the weighted graph of a data structure.\n",
            "Predicted Simplified Sentence: The Middle Is The Seventh Point In The Grain Is The World's Area.\n",
            "\n",
            "Original Sentence: Dijkstra algorithms used for calculating the shortest path, which was introduced by the famous Dutch computer scientist Eiziga.\n",
            "Predicted Simplified Sentence: Can Created For The Short Form And Was Made By The Famous Of Porto Amplifier.\n",
            "\n",
            "Original Sentence: Dijkstra, was recognized as the best algorithm that can be applied to get the shortest path from a node to any other nodes.\n",
            "Predicted Simplified Sentence: Was The Only Uncertain Stance Might That Have A Only Based From The Number Of Greater Services From Other Other Or Users.\n",
            "\n"
          ]
        }
      ],
      "source": [
        "import tensorflow as tf\n",
        "import numpy as np\n",
        "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
        "import pickle\n",
        "\n",
        "# Assuming you have your model loaded, tokenizer, and other configurations\n",
        "\n",
        "# Provided inference sentences\n",
        "inference_sentences = [\n",
        "    \"The shortest path problem based on the data structure has become one of the hot research topics in graph theory.\",\n",
        "    \"As the basic theory of solving this problem, Dijkstra algorithm has been widely used in engineering calculations.\",\n",
        "    \"Aiming at the shortcomings of traditional Dijkstra algorithm, this paper has proposed an optimization method which has mainly improved the nodes selection of the shortest path and data storage structure and organization.\",\n",
        "    \"Through comparison and analysis, the improved algorithm has been obtained, which has reduced the storage space, improved the operational efficiency and has a better applicability in the shortest path calculation.\",\n",
        "    \"People's requirements for living and production efficiency have improved with the rapid development of computer science, in which the shortest path problem has become more prominent, and gradually becomes one of the hot topics of geographic information science, operations research, computer science and other disciplines.\",\n",
        "    \"It is not only the key problem in network analysis but also the key issue in graph theory, and therefore it has been widely used in electronic navigation, network optimization, logistics, transportation and other areas in people's daily life.\",\n",
        "    \"The shortest path is also the one between the nodes in the weighted graph of a data structure.\",\n",
        "    \"Dijkstra algorithms used for calculating the shortest path, which was introduced by the famous Dutch computer scientist Eiziga.\",\n",
        "    \"Dijkstra, was recognized as the best algorithm that can be applied to get the shortest path from a node to any other nodes.\"\n",
        "]\n",
        "\n",
        "# Load the tokenizer used during training\n",
        "with open('tokenizer.pickle', 'rb') as handle:\n",
        "    tokenizer = pickle.load(handle)\n",
        "\n",
        "# Get the index or assign a new one for the '<OOV>' token\n",
        "oov_token_index = tokenizer.word_index.get('<OOV>', None) or len(tokenizer.word_index) + 1\n",
        "\n",
        "# Manually set the index for '<START>' based on your training configuration\n",
        "start_token_index = 1  # Adjust this based on your training data\n",
        "\n",
        "# Placeholder for the results\n",
        "results = []\n",
        "\n",
        "# Loop through inference sentences\n",
        "for sentence in inference_sentences:\n",
        "    # Encode the input sentence\n",
        "    test_input_sequence = tokenizer.texts_to_sequences([sentence])\n",
        "    encoder_test_input_data = pad_sequences(test_input_sequence, maxlen=max_sequence_length, padding='post')\n",
        "    encoder_test_input_data = tf.convert_to_tensor(encoder_test_input_data, dtype=tf.float32)\n",
        "\n",
        "    # Generate predictions using the trained model\n",
        "    decoder_test_input_data = np.zeros((1, max_sequence_length))\n",
        "    decoder_test_input_data[:, 0] = start_token_index  # Manually set the start token index\n",
        "    for i in range(1, max_sequence_length):\n",
        "        predictions = model.predict([encoder_test_input_data, decoder_test_input_data])\n",
        "        predicted_token = np.argmax(predictions[0, i-1, :])\n",
        "        decoder_test_input_data[:, i] = predicted_token\n",
        "        if predicted_token == 0:\n",
        "            continue  # Skip the zero index\n",
        "        if tokenizer.index_word[predicted_token] == '<END>':\n",
        "            break\n",
        "\n",
        "    # Convert predicted sequence to text, replacing OOV tokens\n",
        "    predicted_text = \" \".join([tokenizer.index_word.get(int(idx), '<OOV>') for idx in decoder_test_input_data[0] if idx != 0])\n",
        "\n",
        "    # Append results\n",
        "    results.append({\n",
        "        'original_sentence': sentence,\n",
        "        'predicted_simplified_sentence': predicted_text\n",
        "    })\n",
        "\n",
        "# Display the results\n",
        "for result in results:\n",
        "    # Remove '<OOV>', capitalize the first word, and add a full stop\n",
        "    predicted_simplified_sentence = ' '.join([word.capitalize() for word in result['predicted_simplified_sentence'].split() if word != '<OOV>']) + '.'\n",
        "\n",
        "    print(\"Original Sentence:\", result['original_sentence'])\n",
        "    print(\"Predicted Simplified Sentence:\", predicted_simplified_sentence)\n",
        "    print()\n"
      ]
    }
  ],
  "metadata": {
    "kernelspec": {
      "display_name": "base",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.10.9"
    },
    "colab": {
      "provenance": []
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}