embedding_final_model.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "import platform\n",
    "import random\n",
    "import shutil\n",
    "import sys\n",
    "\n",
    "import h5py\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "import sklearn.metrics\n",
    "import tensorflow as tf\n",
    "from tensorflow.python.saved_model import tag_constants\n",
    "from tqdm import tqdm_notebook as tqdm\n",
    "import keras \n",
    "%matplotlib inline\n",
    "random.seed(0)\n",
    "\n",
    "from vis.visualization import visualize_saliency\n",
    "from vis.utils import utils\n",
    "from keras import activations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def accuracy_plot(model, title): \n",
    "    plt.plot(model.history.history['acc'])\n",
    "    plt.plot(model.history.history['val_acc'])\n",
    "    plt.title(title)\n",
    "    plt.ylabel('Accuracy')\n",
    "    plt.xlabel('Epoch')\n",
    "    plt.legend(['Train', 'Test'], loc='upper left')\n",
    "    plt.show()\n",
    "\n",
    "def loss_plot(model, title): \n",
    "    plt.plot(model.history.history['loss'])\n",
    "    plt.plot(model.history.history['val_loss'])\n",
    "    plt.title(title)\n",
    "    plt.ylabel('Loss')\n",
    "    plt.xlabel('Epoch')\n",
    "    plt.legend(['Train', 'Test'], loc='upper left')\n",
    "    plt.show()\n",
    "    \n",
    "# plotting\n",
    "\n",
    "import itertools\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from sklearn import svm, datasets\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import confusion_matrix\n",
    "from sklearn import metrics\n",
    "\n",
    "def confusion_plot(y_true_labels, y_pred_labels, title, class_names): \n",
    "\n",
    "    def plot_confusion_matrix(cm, classes,\n",
    "                              normalize=False,\n",
    "                              title='Confusion matrix',\n",
    "                              cmap=plt.cm.Blues):\n",
    "        \"\"\"\n",
    "        This function prints and plots the confusion matrix.\n",
    "        Normalization can be applied by setting `normalize=True`.\n",
    "        \"\"\"\n",
    "        if normalize:\n",
    "            cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n",
    "            print(\"All data confusion matrix\")\n",
    "        else:\n",
    "            print('Confusion matrix, without normalization')\n",
    "\n",
    "        print(cm)\n",
    "\n",
    "        plt.imshow(cm, interpolation='nearest', cmap=cmap)\n",
    "        plt.title(title)\n",
    "        plt.colorbar()\n",
    "        tick_marks = np.arange(len(classes))\n",
    "        plt.xticks(tick_marks, classes, rotation=45)\n",
    "        plt.yticks(tick_marks, classes)\n",
    "\n",
    "        fmt = '.2f' if normalize else 'd'\n",
    "        thresh = cm.max() / 2.\n",
    "        for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):\n",
    "            plt.text(j, i, format(cm[i, j], fmt),\n",
    "                     horizontalalignment=\"center\",\n",
    "                     color=\"white\" if cm[i, j] > thresh else \"black\")\n",
    "\n",
    "        plt.tight_layout()\n",
    "        plt.ylabel('True label')\n",
    "        plt.xlabel('Predicted label')\n",
    "\n",
    "    # Compute confusion matrix\n",
    "    cnf_matrix = confusion_matrix(y_true_labels, y_pred_labels)\n",
    "    np.set_printoptions(precision=2)\n",
    "\n",
    "    # Plot non-normalized confusion matrix\n",
    "    plt.figure()\n",
    "    plot_confusion_matrix(cnf_matrix, classes=class_names, normalize=False,\n",
    "                          title=title)\n",
    "\n",
    "    plt.show()\n",
    "    \n",
    "    # Plot normalized confusion matrix\n",
    "    plt.figure()\n",
    "    plot_confusion_matrix(cnf_matrix, classes=class_names, normalize=True,\n",
    "                          title=title)\n",
    "\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>index</th>\n",
       "      <th>Unnamed: 0</th>\n",
       "      <th>Unnamed: 0.1</th>\n",
       "      <th>On-Target Site</th>\n",
       "      <th>Off-Target Site</th>\n",
       "      <th>Score</th>\n",
       "      <th>Encoded Seqs</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>AAATGAGAAGAAGAGGCACAGGG</td>\n",
       "      <td>AAAAGAGAAGCTGAGGCACAGGG</td>\n",
       "      <td>1.0</td>\n",
       "      <td>KKKREKEKKEZREKEEPKPKEEE</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>AAATGAGAAGAAGAGGCACAGGG</td>\n",
       "      <td>AAACGAGAAGAAGAGGCATAGGG</td>\n",
       "      <td>1.0</td>\n",
       "      <td>KKKMEKEKKEKKEKEEPKMKEEE</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>AAATGAGAAGAAGAGGCACAGGG</td>\n",
       "      <td>AAACGAGAAGAAGCTGCACATGG</td>\n",
       "      <td>1.0</td>\n",
       "      <td>KKKMEKEKKEKKEZJEPKPKJEE</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "      <td>AAATGAGAAGAAGAGGCACAGGG</td>\n",
       "      <td>AAATCATATGAAGAGGCACAAGG</td>\n",
       "      <td>1.0</td>\n",
       "      <td>KKKLHKJKREKKEKEEPKPKYEE</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4</td>\n",
       "      <td>5</td>\n",
       "      <td>5</td>\n",
       "      <td>AAATGAGAAGAAGAGGCACAGGG</td>\n",
       "      <td>AAATGAGAAAAATAGGCAGAGGG</td>\n",
       "      <td>1.0</td>\n",
       "      <td>KKKLEKEKKYKKJKEEPKHKEEE</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   index  Unnamed: 0  Unnamed: 0.1           On-Target Site  \\\n",
       "0      0           0             0  AAATGAGAAGAAGAGGCACAGGG   \n",
       "1      1           1             1  AAATGAGAAGAAGAGGCACAGGG   \n",
       "2      2           2             2  AAATGAGAAGAAGAGGCACAGGG   \n",
       "3      3           4             4  AAATGAGAAGAAGAGGCACAGGG   \n",
       "4      4           5             5  AAATGAGAAGAAGAGGCACAGGG   \n",
       "\n",
       "           Off-Target Site  Score             Encoded Seqs  \n",
       "0  AAAAGAGAAGCTGAGGCACAGGG    1.0  KKKREKEKKEZREKEEPKPKEEE  \n",
       "1  AAACGAGAAGAAGAGGCATAGGG    1.0  KKKMEKEKKEKKEKEEPKMKEEE  \n",
       "2  AAACGAGAAGAAGCTGCACATGG    1.0  KKKMEKEKKEKKEZJEPKPKJEE  \n",
       "3  AAATCATATGAAGAGGCACAAGG    1.0  KKKLHKJKREKKEKEEPKPKYEE  \n",
       "4  AAATGAGAAAAATAGGCAGAGGG    1.0  KKKLEKEKKYKKJKEEPKHKEEE  "
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# read in train, val, test dfs \n",
    "data_dir = './data/'\n",
    "# diff sheets, so need to read in \n",
    "file_name = data_dir + 'processed_binary_data' \n",
    "train_df = pd.read_csv(file_name+'_train.csv', sep = ',').reset_index()\n",
    "test_df = pd.read_csv(file_name+'_test.csv', sep = ',').reset_index()\n",
    "train_df.head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Guide length:  23\n",
      "Off-target: 23\n"
     ]
    }
   ],
   "source": [
    "# check lengths -- confirm all are same when comparing across datasets\n",
    "seq_len = len(train_df['On-Target Site'][0])\n",
    "print('Guide length: ', seq_len) # NOTE: WT seq is the reverse complement of guide \n",
    "print('Off-target:', len(train_df['Off-Target Site'][0]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from pysster.One_Hot_Encoder import One_Hot_Encoder\n",
    "\n",
    "from sklearn import preprocessing\n",
    "from sklearn.preprocessing import OneHotEncoder\n",
    "new_alph_map = {'AA': 'K', 'AT': 'R', 'AC': 'Z', 'AG': 'Y', 'TT': 'L', 'TC': 'M', 'TG':'J', \n",
    "                'CC': 'P', 'CG':'H', 'GG': 'E'}\n",
    "\n",
    "new_alph = list(new_alph_map.values())\n",
    "\n",
    "# one-hot encode\n",
    "#modified code from Luis to get correct format for TPOT w/ my alphabet\n",
    "one = One_Hot_Encoder(''.join(new_alph))\n",
    "    \n",
    "def _get_one_hot_encoding(seq):\n",
    "    one_hot_seq = one.encode(seq)                         \n",
    "    encoded_seq = np.argmax((one_hot_seq), axis=1) \n",
    "    return encoded_seq\n",
    "    #return pd.get_dummies(\n",
    "      #list(seq) + new_alph)[:-len(new_alph)].as_matrix().transpose()\n",
    "\n",
    "# now convert the data into one_hot_encoding \n",
    "input_col_name = 'Encoded Seqs'\n",
    "train_data = np.stack(\n",
    "    [_get_one_hot_encoding(s) for s in train_df[input_col_name]]).astype(np.float32)\n",
    "#val_data = np.stack(\n",
    "    #[_get_one_hot_encoding(s) for s in val_df[input_col_name]]).astype(np.float32)\n",
    "test_data = np.stack(\n",
    "    [_get_one_hot_encoding(s) for s in test_df[input_col_name]]).astype(np.float32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "done\n"
     ]
    }
   ],
   "source": [
    "alph_len = len(new_alph)\n",
    "\n",
    "from keras.utils import to_categorical\n",
    "\n",
    "output_bin_col_name = 'Score'; \n",
    "\n",
    "# NOTE: need one-hot encoding of the labels \n",
    "train_bins = to_categorical(np.array(train_df[output_bin_col_name].astype(np.float32)))\n",
    "#val_bins = to_categorical(np.array(val_df[output_bin_col_name].astype(np.float32)))\n",
    "test_bins = to_categorical(np.array(test_df[output_bin_col_name].astype(np.float32)))\n",
    "\n",
    "print('done')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.layers.core import Permute\n",
    "from keras import backend as K\n",
    "from keras.engine.topology import Layer\n",
    "import numpy as np\n",
    "import keras as keras\n",
    "from keras.callbacks import TensorBoard\n",
    "from keras import metrics as metrics\n",
    "from keras.models import Sequential, Model\n",
    "from keras.layers import Dense, Dropout, Activation, Flatten, Input, Embedding\n",
    "from keras.optimizers import SGD\n",
    "from keras.regularizers import l2\n",
    "\n",
    "def embed_model(embed_dim, hidden_layers, l2_coeff, dropout_rate):\n",
    "    X_in = Input(shape=(seq_len,),dtype='float32')\n",
    "    E = Embedding(alph_len,embed_dim,input_length=seq_len)(X_in)\n",
    "    H = Flatten()(E)\n",
    "    H = Dropout(dropout_rate)(H)\n",
    "    for h in hidden_layers: \n",
    "        H = Dense(h, activation='relu', activity_regularizer = l2(l2_coeff))(H)\n",
    "        H = Dropout(dropout_rate)(H)\n",
    "    H = Dense(2,activation=\"softmax\",name='preds')(H)\n",
    "    model = Model(inputs=[X_in], outputs=H)\n",
    "    model.compile(loss='binary_crossentropy',optimizer='adam',metrics=['acc'])#[precision,recall])#metrics=[sensitivity, specificity])\n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "327078/327078 [==============================] - 3s 9us/step\n",
      "Epoch 0: Loss  0.6761063544646078  accuracy  0.9002623227483377\n",
      "Epoch 1/20\n",
      "327078/327078 [==============================] - 3s 8us/step - loss: 0.0715 - acc: 0.9982\n",
      "Epoch 2/20\n",
      "327078/327078 [==============================] - 2s 7us/step - loss: 0.0188 - acc: 0.9983A: 1s\n",
      "Epoch 3/20\n",
      "327078/327078 [==============================] - 2s 7us/step - loss: 0.0098 - acc: 0.9987\n",
      "Epoch 4/20\n",
      "327078/327078 [==============================] - 2s 7us/step - loss: 0.0070 - acc: 0.9987A: 0s - loss: 0.0072 -\n",
      "Epoch 5/20\n",
      "327078/327078 [==============================] - 2s 7us/step - loss: 0.0060 - acc: 0.9987\n",
      "Epoch 6/20\n",
      "327078/327078 [==============================] - 2s 7us/step - loss: 0.0056 - acc: 0.9988\n",
      "Epoch 7/20\n",
      "327078/327078 [==============================] - 2s 7us/step - loss: 0.0053 - acc: 0.9988\n",
      "Epoch 8/20\n",
      "327078/327078 [==============================] - 2s 7us/step - loss: 0.0051 - acc: 0.9988\n",
      "Epoch 9/20\n",
      "327078/327078 [==============================] - 2s 7us/step - loss: 0.0050 - acc: 0.9988\n",
      "Epoch 10/20\n",
      "327078/327078 [==============================] - 2s 7us/step - loss: 0.0048 - acc: 0.9989A: 0s - loss: 0.0047 - acc: 0.99\n",
      "Epoch 11/20\n",
      "327078/327078 [==============================] - 2s 7us/step - loss: 0.0047 - acc: 0.9988\n",
      "Epoch 12/20\n",
      "327078/327078 [==============================] - 2s 7us/step - loss: 0.0045 - acc: 0.9989\n",
      "Epoch 13/20\n",
      "327078/327078 [==============================] - 3s 8us/step - loss: 0.0045 - acc: 0.9989\n",
      "Epoch 14/20\n",
      "327078/327078 [==============================] - 3s 8us/step - loss: 0.0044 - acc: 0.9989\n",
      "Epoch 15/20\n",
      "327078/327078 [==============================] - 3s 9us/step - loss: 0.0042 - acc: 0.9990\n",
      "Epoch 16/20\n",
      "327078/327078 [==============================] - 3s 9us/step - loss: 0.0042 - acc: 0.9990\n",
      "Epoch 17/20\n",
      "327078/327078 [==============================] - 3s 9us/step - loss: 0.0040 - acc: 0.9990A: 0s - loss: 0.0039 - acc: 0\n",
      "Epoch 18/20\n",
      "327078/327078 [==============================] - 3s 9us/step - loss: 0.0040 - acc: 0.9990A: 1s - loss: 0.0042 - ETA: 0s - loss: 0.0040 \n",
      "Epoch 19/20\n",
      "327078/327078 [==============================] - 3s 9us/step - loss: 0.0039 - acc: 0.9990A:  - ETA: 0s - loss: 0.0038 \n",
      "Epoch 20/20\n",
      "327078/327078 [==============================] - 3s 9us/step - loss: 0.0038 - acc: 0.9990A: 0s - loss: 0.00\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x125fe7780>"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = embed_model(embed_dim=5, hidden_layers=(50,), l2_coeff = 0.001, dropout_rate = 0.0)\n",
    "\n",
    "X = train_data\n",
    "y = train_bins\n",
    "num_epochs = 20\n",
    "#early_stopping = keras.callbacks.EarlyStopping(monitor='val_loss', min_delta=0.005, patience=10, verbose=0, mode='auto', baseline=None)\n",
    "initial_loss, initial_accuracy = model.evaluate(X, y)\n",
    "print(\"Epoch 0: Loss \", initial_loss, \" accuracy \", initial_accuracy)\n",
    "model.fit(X, y, epochs=num_epochs, batch_size=128)#,validation_data=(val_data, val_bins))#,callbacks=[early_stopping])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AUC: 0.7615506490777929\n",
      "Accuracy: 0.998584053513012 \n",
      "\n",
      "F1: 0.8023664654973843 \n",
      "\n",
      "Confusion matrix, without normalization\n",
      "[[81719    35]\n",
      " [   81    89]]\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "All data confusion matrix\n",
      "[[1.00e+00 4.28e-04]\n",
      " [4.76e-01 5.24e-01]]\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Average precision-recall score: 0.84\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1.0, '2-class Precision-Recall curve: AP=0.84')"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "specs = 'Alphabet Embedding'\n",
    "\n",
    "X_new = test_data\n",
    "y_true = test_bins\n",
    "y_pred = model.predict(X_new)\n",
    "\n",
    "y_true_labels = np.argmax(y_true, axis=1)\n",
    "y_pred_labels = np.argmax(y_pred, axis=1) \n",
    "from sklearn.metrics import roc_auc_score\n",
    "\n",
    "print('AUC:', roc_auc_score(y_true_labels, y_pred_labels))\n",
    "\n",
    "print('Accuracy:', sklearn.metrics.accuracy_score(y_true_labels, y_pred_labels), '\\n')\n",
    "print('F1:', sklearn.metrics.f1_score(y_true_labels,y_pred_labels, average='macro'),'\\n')\n",
    "\n",
    "#accuracy_plot(model, 'Accuracy: ' + specs)\n",
    "#loss_plot(model, 'Loss: ' + specs)\n",
    "\n",
    "class_names = ['0', '1']\n",
    "confusion_plot(y_true_labels, y_pred_labels, 'Confusion Matrix: ' + specs, class_names)\n",
    "\n",
    "# from TAs on piazza: use precision-recall, ROC can be misleading \n",
    "# modified code from sklearn documentation: https://scikit-learn.org/stable/modules/generated/sklearn.metrics.precision_recall_curve.html\n",
    "# and referenced: https://machinelearningmastery.com/roc-curves-and-precision-recall-curves-for-classification-in-python/\n",
    "from sklearn.metrics import precision_recall_curve, average_precision_score\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.utils.fixes import signature\n",
    "\n",
    "average_precision = average_precision_score(y_true, y_pred)\n",
    "\n",
    "print('Average precision-recall score: {0:0.2f}'.format(\n",
    "      average_precision))\n",
    "\n",
    "precision, recall, _ = precision_recall_curve(y_true_labels, y_pred[:,1]) # look at positive class prob\n",
    "\n",
    "# In matplotlib < 1.5, plt.fill_between does not have a 'step' argument\n",
    "step_kwargs = ({'step': 'post'}\n",
    "               if 'step' in signature(plt.fill_between).parameters\n",
    "               else {})\n",
    "plt.step(recall, precision, color='b', alpha=0.2,\n",
    "         where='post')\n",
    "plt.fill_between(recall, precision, alpha=0.2, color='b', **step_kwargs)\n",
    "\n",
    "plt.xlabel('Recall')\n",
    "plt.ylabel('Precision')\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlim([0.0, 1.0])\n",
    "plt.title('2-class Precision-Recall curve: AP={0:0.2f}'.format(\n",
    "          average_precision))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[<keras.engine.input_layer.InputLayer object at 0x129e67048>, <keras.layers.embeddings.Embedding object at 0x129e670b8>, <keras.layers.core.Flatten object at 0x129e67160>, <keras.layers.core.Dropout object at 0x129e67198>, <keras.layers.core.Dense object at 0x129e671d0>, <keras.layers.core.Dropout object at 0x129e672b0>, <keras.layers.core.Dense object at 0x129e67470>]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<seaborn.matrix.ClusterGrid at 0x13fc729e8>"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# look at the learned weights to viz embedding of new alphabet\n",
    "weight_matrix = model.layers[1].get_weights()[0]\n",
    "weights = pd.DataFrame(weight_matrix)\n",
    "print(model.layers)\n",
    "import seaborn as sns\n",
    "reverse_map = {v:k for k,v in new_alph_map.items()}\n",
    "sns.clustermap(weights,yticklabels=[reverse_map[letter] for letter in new_alph])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1.0, 'PCA over Learned Embedding Weights')"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAWEAAAFNCAYAAADRkd6aAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvOIA7rQAAIABJREFUeJzt3XucVXW9//HX2wEExUSBFFAuKngSDJVRw8xULMwLQppRHRvPsfhRlnn6aeEvT5Ja6aFf0Sm7WHa0c5LJTJCfliSCohTpoMilRC5eYIARQVQU5dLn98deM26GmWEG997fYeb9fDz2Y9b6ru/a67P37HnP2t+191qKCMzMLI19UhdgZtaeOYTNzBJyCJuZJeQQNjNLyCFsZpaQQ9jMLCGHsLUbkp6XdFYJtnOppMcKdF/9JYWkDo0snyjpf7LpvpI2SyorxLaLRdJnJP2pmX0L9ly2Vg7hAsr+yLdkfwg1km6X1DVv+UhJcyS9Lmm9pEckjap3H6dnf3RfL/0jaJ7scd2Yuo5Cyh7T1ux3V3t7OnVdLRERL0ZE14jYUcj7lTQ8e82W5bX9opG2nzWjzt9ExEcLVNvDkj5XiPtKxSFceOdHRFfgBKAcuBZA0kXA74BfA4cBhwDfBM6vt34FsBH4bKkKbkpje2BtdLv/kYVY7W1oghpaoypyWXFCXtuHgNX12k4D5pSwrjbBIVwkEVEN/BEYIknA94EbIuKXEfFqRPwjIh6JiM/XriNpf+Ai4HJgoKTyprYh6fOSlkvaKGm6pN5Z+08lfa9e33slfTWb7i3p99ne+HOSrsjrN1HS3ZL+R9JrwKUtedyS/knSg1lNSyVdnLfsXElPSXpN0ipJE/OW1b7tvkzSi8CsvLYKSS9KelnSN/LW2UfSBEkrJG2QdJekg/OWXyLphWxZ3XotlVfHv2R1vyJpvKQTJS2UtEnSj3ddTT+W9KqkZySNyFtwoKTbJK2VVC3pxto9Skllkr6XPdaVwLn17nRA9g7qdUkPAj0aqLNDNv+wpBskzc36/0lSfv/P5j0//65GhmsiYhswj1zIIum9QCfgrnptg8hCeDePcachBkkfzV4rr0r6Sfb4dtq7zZ6TV7LX68eytm+T+2fwY+XeufxYOT+Q9FL2Olskacjuf8sJRYRvBboBzwNnZdOHA0uAG4B/AgIYsJv1LwHWAmXA/wN+1ETfM4GXye2J7Av8CJiTLTsNWAUomz8I2AL0JvePdz65vfBOwBHASmBk1ncisA0YnfXt0sC2bwdubKB9/2y7/wJ0AI7PajwmW346cGx2v+8HaoDR2bL+2XP06+x+uuS1/SKbHwq8DbwvW+cr5MLhsOw5+DkwJVt2DLA5ey72JfdPcHvt76e5j6lebT8DOgMfBd4CpgHvBfoALwEfzvpfmm3r34COwCeBV4GDs+VTs1r3z9Z/HPhf2bLxwDPZ6+dgYHa27Q7Z8r9kj2Xf7LG9DvxPvTpr+z4MrCAXjl2y+ZvqPT+nZq+D72W/98aen+uAe7Ppi7Lf00fqta3M69/UY7wUeCyb7gG8Bnyc3GvmK1kdn8vruw34PLm/iy8Aa3jntf1wbd9sfiS513c3QMD7gF6ps6HJv/vUBbSlG7kQ3gxsAl4AfpK9+D+Y/XF03s36M4HJ2fSngPVAx0b63kbu7XPtfNfsxdo/e/G9CJyWLfs8MCubPhl4sd59XQP8VzY9kSzMm6jzdhoO4U8Cj9Zr+zlwXSP3Mxn4QTZdGyBH5C2vbTssr+1xYGw2/XdgRN6yXtlz0IHcP5nKvGX7A1tpOoTfyn53tbc76tXRJ6//BuCTefO/B67Mpi/ND4q8ui8hNwz1Nnn/3LLf9exsehYwPm/ZR7NtdwD6kgv3/fOW30nTIXxtXt8vAg9k098k+4eVze+3m+fn9OwxC/hh9prqSu4faW1b7Wtod4/xUt4J4c8Cf8nrJ3L/yPNDeHm9OgM4NO8x5ofwmcCzwAeAfUqdAXtySzLe18aNjoiZ+Q2SNmSTvYDnGlpJ0uHAGeQCEeBe4FZyb0enNbBKb+DJ2pmI2Jxtp09EPC+pktwLfw7waeB/sq79gN6SNuXdVxnwaN78qt09yEb0A06ud98dgP8GkHQycBMwhNze177kxsnzNbTtdXnTb5L746/d3lRJ/8hbvoNcCPTOv6+IeCPv99CY70XEtU0sr8mb3tLAfNe8+erIUiHzQlZTP3J7x2sl1S7bJ6/WnerO1iNv2SsR8Ua95Yc3UXNjz1395+fN3Tw/87J1h5DbA/9p9ppbldf2n1nf3T3GfPXrCEmrG3sMWZ2w83NN3vJZ2dDQLUA/SfcAV0XEa008tqQ8JlwaS8m90C5sos8l5H4f/0/SOnJDBJ3JHahryBpyL3agbjy5O1CdNU0BLpLUj9ze7++z9lXAcxHRLe92QESck3ffe3pqvVXAI/Xuu2tEfCFbficwHTg8Ig4k9/Ze9e6jJdteBXys3vY6R248fi154SRpP3LPT6n0UV4CkduLXZPV/DbQI6/m90TE4KzfTnVn65G37KDsd93Q8pZYS24YBwBJXWji+YmIt4AnyB1I7hURz2SLHs3a3s87B+V29xibqkP5882wy+slIv4zIoaRG3IZBFzdgvsrOYdwCWR7RF8F/j07uPOe7KDSqZJuzbpVAN8Cjsu7XQicI6mhP44pwL9IOk7SvsB3gL9GxPPZNp8iNx77S2BGRNTunT4OvC7p65K6ZAeChkg6sYUPq0xS57xbJ+A+YJByB8Q6ZrcTJb0vW+cAYGNEvCXpJHJ76O/Gz4BvZ/9okNRT0gXZsruB87LnuBNwPaV9vb8XuCJ7Dj5BbmzyDxGxFvgT8H/zXgdHSvpwtt5d2XqHSToImFB7hxHxArlPKnxLUidJp7Lrp2ua627gfEmnZM/PRHb9h1jfHHJjtn/Oa3ssa1sbESuyOnf3GPPdDxwraXR2QPFy4NAWPI4acsc1AMhebydL6gi8QW6I6R+NrdwaOIRLJCLuJjdm+q/k9ohqgBuBeyV9gNxe7S0RsS7vNh1YTm5Yof79zQT+ndwe7lrgSGBsvW53AmdlP2vX2wGcRy7kn+OdoD6whQ9pArm34LW3WRHxOrkxzLHZY1wH3Exu2AFyY5LXS3qd3JjkXS3cZn0/JLdn/afsPueR2+snIpaQ+4O+k9zz8wq5j1Q15Wva+XPCL7+L2v4KDCT3/H4buCgiat/uf5bccMzfsrruJjdUBbmDkDOAp8kNN91T734/Te4xbiR3sOzXe1Jc9vx8Gagk9/xsJndw8e0mVnuE3D+X/C9PPJa1PVqvb1OPMb+Ol4FPAP9Bbsz5GHL/aJqqI98Pyb3je0XSfwLvIfccvkJuqGYDMKmZ95VE7RFGM2vHlPtS0SZgYEQ0eNyiRHXsQ+6f5WciYnaqOkrJe8Jm7ZSk8yXtl40xfw9YRO4TPqWuY6Skbtmw2v8hNywyr9R1pOIQNmu/LiA3bLSG3NDJ2Ejz1ng4uc8zv0xujHt0RGxJUEcSHo4wM0vIe8JmZgk5hM3MEmpX35jr0aNH9O/fP3UZZtbGzJ8//+WI6Lkn67arEO7fvz9VVVWpyzCzNkbSC7vv1TAPR5iZJeQQNjNLyCFsZpaQQ9jMLCGHsJlZQg5hM7OEHMJmZgk5hM3MEnIIm5kl1K6+MWfWWm3YsIERI0YAsG7dOsrKyujZM/ct2OnTp/O1r32NJ554gm7dunHIIYcwefJkBg0alLJkKxCHsFkr0L17dxYsWADAxIkT6dq1K1dddRURwSmnnEJFRQWVlZUAPP3009TU1DiE2wiHsFkrNnv2bDp27Mj48ePr2oYOHZqwIis0jwmbtWKLFy9m2LBhqcuwIkoawpLOlrRU0nJJExpY/lVJf5O0UNJDtZc2z5btkLQgu00vbeVmZoWRbDhCUhlwC/ARcldXfULS9Ij4W163p4DyiHhT0hfIXRb7k9myLRFxXEmLNiuwaU9VM2nGUtZs2kLvbl24euTROy0fPHgwd999d6LqrBRS7gmfBCyPiJURsRWoJHfhwToRMTsi3sxm5wGHlbhGs6KZ9lQ119yziOpNWwigetMWrrlnEc+sfa2uz5lnnsnbb7/NrbfeWte2cOFCHn300QQVWzGkDOE+wKq8+dVZW2MuA/6YN99ZUpWkeZJGF6NAs2KaNGMpW7bt2Klty7YdzF2xoW5eElOnTmXmzJkceeSRDB48mGuuuYZDDz201OVakewVn46Q9M9AOfDhvOZ+EVEt6QhglqRFEbGigXXHAeMA+vbtW5J6zZpjzaaGr+reofxirrrq3Lr53r17c9ddd5WqLCuxlHvC1cDhefOHZW07kXQW8A1gVES8XdseEdXZz5XAw8DxDW0kIm6NiPKIKK/98LtZa9C7W5cWtVvblDKEnwAGShogqRMwFtjpUw6Sjgd+Ti6AX8prP0jSvtl0D+CDQP4BPbNW7+qRR9OlY9lObV06lu1ycM7atmTDERGxXdKXgBlAGfCriFgi6XqgKiKmA5OArsDvJAG8GBGjgPcBP5f0D3L/SG6q96kKs1Zv9PG5QyD1Px1R227tgyIidQ0lU15eHr7aspkVmqT5EVG+J+v6G3NmZgk5hM3MEnIIm5kl5BA2M0vIIWxmlpBD2MwsIYewmVlCDmEzs4QcwmZmCTmEzcwScgibmSXkEDYzS8ghbGaWkEPYzCwhh7CZWUIOYTOzhBzCZmYJOYTNzBJyCJuZJeQQNjNLyCFsZpaQQ9jMLKGkISzpbElLJS2XNKGB5V+V9DdJCyU9JKlf3rIKScuyW0VpKzczK4xkISypDLgF+BhwDPApScfU6/YUUB4R7wfuBv4jW/dg4DrgZOAk4DpJB5WqdjOzQkm5J3wSsDwiVkbEVqASuCC/Q0TMjog3s9l5wGHZ9EjgwYjYGBGvAA8CZ5eobjOzgkkZwn2AVXnzq7O2xlwG/LGl60oaJ6lKUtX69evfRblmZoW3VxyYk/TPQDkwqaXrRsStEVEeEeU9e/YsfHFmZu9CyhCuBg7Pmz8sa9uJpLOAbwCjIuLtlqxrZtbapQzhJ4CBkgZI6gSMBabnd5B0PPBzcgH8Ut6iGcBHJR2UHZD7aNZmZrZX6ZBqwxGxXdKXyIVnGfCriFgi6XqgKiKmkxt+6Ar8ThLAixExKiI2SrqBXJADXB8RGxM8DDOzd0URkbqGkikvL4+qqqrUZZhZGyNpfkSU78m6e8WBOTOztsohbGaWkEPYzCwhh7CZWUIOYTOzhBzCZmYJOYTNzBJyCJuZJeQQNjNLyCFsZpaQQ9jMLCGHsJlZQg5hM7OEHMJmZgk5hM3MEnIIm5kl5BA2M0vIIWxmlpBD2MwsIYewmVlCDmEzs4SShrCksyUtlbRc0oQGlp8m6UlJ2yVdVG/ZDkkLstv00lVtZlY4HVJtWFIZcAvwEWA18ISk6RHxt7xuLwKXAlc1cBdbIuK4ohdqZlZEyUIYOAlYHhErASRVAhcAdSEcEc9ny/6RokAzs2JLORzRB1iVN786a2uuzpKqJM2TNLqwpZmZlUbKPeF3q19EVEs6ApglaVFErKjfSdI4YBxA3759S12jmVmTUu4JVwOH580flrU1S0RUZz9XAg8DxzfS79aIKI+I8p49e+55tWZmRZAyhJ8ABkoaIKkTMBZo1qccJB0kad9sugfwQfLGks3M9hbJQjgitgNfAmYAfwfuioglkq6XNApA0omSVgOfAH4uaUm2+vuAKklPA7OBm+p9qsLMbK+giEhdQ8mUl5dHVVVV6jLMrI2RND8iyvdkXX9jzswsIYewmVlCDmEzs4QcwmZmCTmEzcwScgibmSXkEDYzS8ghbGaWkEPYzCwhh7CZWUIOYTOzhBzCZmYJOYTNzBJyCJuZJeQQNjNLyCFsZpaQQ9jMLCGHsJlZQg5hM7OEHMJmZgk5hM3MEkoawpLOlrRU0nJJExpYfpqkJyVtl3RRvWUVkpZlt4rSVW1mVjjJQlhSGXAL8DHgGOBTko6p1+1F4FLgznrrHgxcB5wMnARcJ+mgYtdsZlZoKfeETwKWR8TKiNgKVAIX5HeIiOcjYiHwj3rrjgQejIiNEfEK8CBwdimKNjMrpJQh3AdYlTe/Omsr9rpmZq1Gmz8wJ2mcpCpJVevXr09djpnZTlKGcDVweN78YVlbQdeNiFsjojwiynv27LlHhZqZFUvKEH4CGChpgKROwFhgejPXnQF8VNJB2QG5j2ZtZmZ7lWQhHBHbgS+RC8+/A3dFxBJJ10saBSDpREmrgU8AP5e0JFt3I3ADuSB/Arg+azMz26soIlLXUDLl5eVRVVWVugwza2MkzY+I8j1Zt80fmDMza80cwmZmCTmEzcwScgibmSXkEDYzS8ghbGaWkEPYzCwhh7CZWUIOYTOzhBzCZmYJOYTNzBJqMoQl/ZOkEZK61mv3VSzMzAqg0RCWdAVwL/BlYLGk/EsPfafYhZmZldK6desYO3YsRx55JMOGDeOcc87h2WefZdmyZZx33nl17WeccQZz5swp2HY7NLHs88CwiNgsqT9wt6T+EfFDQAWrwMwssYhgzJgxVFRUUFlZCcDTTz9NTU0Nl112Gd/73vcYNWoUAIsXL6aqqorTTjutINtuKoT3iYjNWYHPSzqdXBD3wyFsZm3I7Nmz6dixI+PHj69rGzp0KLfddhvDhw+vC2CAIUOGMGTIkIJtu6kx4RpJx9XOZIF8HtADOLZgFZiZJbZ48WKGDRu2S/uSJUs44YQTirrtpvaEPwtsz2/IrobxWUk/L2pVZmZFNu2paibNWMqaTVvQkmcZ8p63d7vOmDFjWLZsGYMGDeKee+4pSB2N7glHxOqIWNfIsrkF2bqZWQLTnqrmmnsWUb1pCwG8sX8vZj46j2lP7Xy94MGDB/Pkk0/WzU+dOpXbb7+djRsLdzU1f07YzNqdSTOWsmXbjrr5zv2GsmP7Vv73Df+3rm3hwoUMGjSIuXPnMn36O9cgfvPNNwtaS1PDEWZmbdKaTVt2mpdEzzHfYN1Dv+DII4+kc+fO9O/fn8mTJ3Pffffx1a9+lSuvvJJDDjmEAw44gGuvvbZgtTQawpKOAg6pP/Qg6YPAuohYUbAqzMxKqHe3LlTXC+IOB3TnuEu/xdwJZ+7S/w9/+EPRamlqOGIy8FoD7a9ly8zM9kpXjzyaLh3Ldmrr0rGMq0ceXfJamgrhQyJiUf3GrK1/ITYu6WxJSyUtlzShgeX7Svpttvyv2ZdGkNRf0hZJC7LbzwpRj5m1D6OP78N3P34sfbp1QUCfbl347sePZfTxfUpeS1Njwt2aWNbl3W5YUhlwC/ARYDXwhKTpEfG3vG6XAa9ExFGSxgI3A5/Mlq2IiOMwM9sDo4/vkyR062tqT7hK0ufrN0r6HDC/ANs+CVgeESsjYitQCVxQr88FwB3Z9N3ACEn+tp6ZtRlN7QlfCUyV9BneCd1yoBMwpgDb7gOsyptfDZzcWJ+I2C7pVaB7tmyApKfIjVFfGxGPFqAmM7OSajSEI6IGOEXSGUDtF6Xvj4hZJamsaWuBvhGxQdIwYJqkwRGxy4FESeOAcQB9+/YtcZlmZk1r6lSWnSVdCVwIbAV+WuAArgYOz5s/LGtrsI+kDsCBwIaIeDsiNgBExHxgBTCooY1ExK0RUR4R5T179ixg+WZm715TY8J3kBt+WAR8DPhegbf9BDBQ0gBJnYCxwPR6faYDFdn0RcCsiAhJPbMDe0g6AhgIrCxwfWZmRdfUmPAxEXEsgKTbgMcLueFsjPdLwAygDPhVRCyRdD1QFRHTgduA/5a0HNhILqgBTgOul7QN+AcwPiIK92VuM7MSaSqEt9VOZIFZ8I1HxB+AP9Rr+2be9FvAJxpY7/fA7wtekJlZiTUVwkMl1R7oEtAlmxcQEfGeoldnZtbGNfXpiLLGlpmZWWH4VJZmZgk5hM3MEnIIm5kl5BA2K6Fp06YhiWeeeaaubdmyZZx33nkceeSRDBs2jDPOOIM5c+YkrNJKySFsVkJTpkzh1FNPZcqUKQC89dZbnHvuuYwbN44VK1Ywf/58fvSjH7Fypb971F44hM1KZPPmzTz22GPcdtttVFZWAvCb3/yG4cOHM2rUqLp+Q4YM4dJLL01UpZWaQ7gAGnqLCTB58mQ6d+7Mq6++mqgya03uvfdezj77bAYNGkT37t2ZP38+S5Ys4YQTTkhdmiXkEC6A+m8x89tPPPFE7rnnnkSVWWsyZcoUxo7NffN+7Nixu7xeAMaMGcOQIUP4+Mc/XuryLBGH8LvU0FtMgBUrVrB582ZuvPHGBv/YrH2Y9lQ1H7xpFn2/Uskf/zSTz3z2X+jfvz+TJk3irrvuYvDgwTz55JN1/adOncrtt9/Oxo0+FUp74RB+lxp6iwlQWVnJ2LFj+dCHPsTSpUupqalJXKmV2rSnqrnmnkVUb9rCG0vnst8xZ9D9c79k8tS5rFq1igEDBnDUUUcxd+5cpk9/5wSCb775ZsKqrdQcwu9SY28xa9v32WcfLrzwQn73u9+lLNMSmDRjKVu27QDgjb8/wn6DhrNl2w4mzVgKwIUXXkhlZSX33XcfP/vZzzjiiCMYPnw4N954I9dee23K0q2EFBGpayiZ8vLyqKqqetf3M+2paibNWMqqtS9R/dNL6d69B/vt24EdO3Ygifvvv5/y8nJ69eoFwNatWxkwYABz585919u2vceACffT0F+XgOduOrfU5VgRSZofEeV7sq73hFuoOW8xv/KVrzBx4kSef/55nn/+edasWcOaNWt44YUXUpdvJdS7W8MXJW+s3donh3ALNect5nPPPceYMTtfC3XMmDE7Hbiztu/qkUfTpePOJyPs0rGMq0cenagia408HNFCfotpLVE7dLVm0xZ6d+vC1SOPZvTxfVKXZQX2boYjmjqpuzWgd7cuVG/a0mC7WX2jj+/T6kJ32rRpjBkzhr///e9s27aNSy65BIAXX3yRAw88kAMPPJAePXowc+bMxJW2Dx6OaCG/xbS9Xf6Xi4499lgWLFjAggULGDVqFJMmTWLBggUO4BJyCLfQ6OP78N2PH0ufbl0Q0KdbF7778WNb3d6OWUMa+3KRpePhiD3QGt9imjVHQ18uGjZsWOqy2rWke8KSzpa0VNJySRMaWL6vpN9my/8qqX/esmuy9qWSRpaybrO9Se1XpwdMuJ8v3/AjDi8/C2j8/BVWWslCWFIZcAvwMeAY4FOSjqnX7TLglYg4CvgBcHO27jHAWGAwcDbwk+z+Wr2amho+/elPc8QRRzBs2DCGDx/O1KlTAXj88cc5/fTTGThwICeccALnnnsuixYtSlyx7c3yP9e+fcvrbFqxgG9PuJL39j687vwV7ekTUq1Ryj3hk4DlEbEyIrYClcAF9fpcANyRTd8NjJCkrL0yIt6OiOeA5dn9tWoRwejRoznttNNYuXIl8+fPp7KyktWrV1NTU8PFF1/Md77zHZYtW8aTTz7JNddcw4oVK1KXbXux/M+1v7l0LvsPPoM+X/gVA6+4o+7LRY8++mjiKtu3lCHcB1iVN786a2uwT0RsB14Fujdz3VZn1qxZdOrUifHjx9e19evXjy9/+cv8+Mc/pqKiglNOOaVu2amnnsro0aNTlGptxJq8j1PWfrkov/3CCy/0kERibf7AnKRxwDiAvn37Jq2lqRN4L1myhIqKihJXZG1d/ufaD/3Ud3dqB7jiiivq2m6//faS1mY5KfeEq4HD8+YPy9oa7COpA3AgsKGZ6wIQEbdGRHlElPfs2bNApTdf/kGRyQ8+y4qXNtctu/zyyxk6dCgnnnjiLuudfPLJvO997+MrX/lKKcu1Nsafa2/9UobwE8BASQMkdSJ3oG16vT7Tgdrdw4uAWZE7ijAdGJt9emIAMBB4vER1N1v+QZEA3ti/FzMfm8e0p3L/L2655RYeeugh1q9fv8vJvf/6179yww03+NJI9q74c+2tX7LhiIjYLulLwAygDPhVRCyRdD1QFRHTgduA/5a0HNhILqjJ+t0F/A3YDlweETuSPJAm5B8UAejcbyib5vya//2tSYyeNhl45wTel19+OSeffDIjR46sGxf2yb2tEPy59tbNJ/ApooZO9rN980ZeeegXvPetVfTs2ZP999+f8ePH88lPfpJ58+bx9a9/nerqat773vfSo0cPvvnNb1JevkfnBTGzEvEJfFqphk7206HrwRxXMZG5E87cpf8HPvABHnnkkVKVZ2atgM8dUUQ+KGJmu+M94SKqHYfz+WTNrDEO4SLzQREza4qHI8zMEnIIm5kl5BA2M0vIIWxmlpBDeA+sW7eOsWPHcuSRRzJs2DDOOeccnn32WZ599lnOOeecuvMBX3zxxdTU1KQu18xaMX86ooUigjFjxlBRUVF3ja6nn36ampoa/vVf/5Xvf//7nH/++QA8/PDDrF+/nkMOOSRlyWbWinlPuIVmz55Nx44ddzon8NChQ1m2bBnDhw+vC2CA008/nSFDhqQo08z2Eg7hFlq8eHGDF0ZsrN3MrCkOYTOzhDwm3AzTnqqu++rxfi9vhSf/skufwYMH++Q7ZtZi3hPejfonZt/c/Z9YWbOJL157c12fhQsXMmjQIP785z9z//3317XPmTOHxYsXJ6jazPYWPp/wbnzwplm7nI5y++sbeOvRX3HgG6vo3Lkz/fv3Z/LkyezYsYMrr7ySFStW0LFjR97//vfzwx/+0J+OMGvjfD7hIlpTL4ABOhzQnQPOuZoVN527y7IHHnigFGWZWRvh4YjdqL0qbXPbzcxawiG8Gz4xu5kVk4cjdsMnZjezYnIIN4NPzG5mxeLhCDOzhJKEsKSDJT0oaVn286BG+lVkfZZJqshrf1jSUkkLstt7S1e9mVnhpNoTngA8FBEDgYey+Z1IOhi4DjgZOAm4rl5YfyYijstuL5WiaDOzQks1JnwBcHo2fQfwMPD1en1GAg9GxEYASQ8CZwNTSlNi4WzYsIERI0YAuXMRl5WVcfDBB1NWVsbWrVvZuHEjZWVl9OzZE4DHH3+cTp06pSzZzEokVQgfEhFrs+l1QENfKesDrMqbX5211fovSTuA3wM3RiNf/ZM0DhgH0Ldv33db9x7p3r07CxYsAGDixIl07dqVq666qm55Q2136ekEAAAMb0lEQVRm1j4ULYQlzQQObWDRN/JnIiIktfS705+JiGpJB5AL4UuAXzfUMSJuBW6F3NeWW7gdM7OiKloIR8RZjS2TVCOpV0SsldQLaGhMt5p3hiwADiM3bEFEVGc/X5d0J7kx4wZD2MysNUs1HDEdqABuyn7e20CfGcB38g7GfRS4RlIHoFtEvCypI3AeMLMENbdY/ikwa7/kYWaWL9WnI24CPiJpGXBWNo+kckm/BMgOyN0APJHdrs/a9gVmSFoILCC3x/yL0j+EptU/BWb1pi1cc88inln7WurSzKwVSbInHBEbgBENtFcBn8ub/xXwq3p93gBa/XWEJs1YypZtO3Zq27JtB3NXbKB8YO9EVZlZa+NvzBVJQ6fABHj9rW0lrsTMWjOfO6JIenfrssvJ4AEGn/c5rrrqzJ3aJk6cWKKqzKy18Z5wkfgUmGbWHN4TLhKfAtPMmsMhXEQ+BaaZ7Y6HI8zMEnIIm5kl5BA2M0vIIWxmlpBD2MwsIYewmVlCDmEzs4QcwmZmCTmEzcwScgibmSXkEDYzS8ghbGaWkEPYzCwhh7CZWUIOYTOzhBzCZmYJJQlhSQdLelDSsuznQY30e0DSJkn31WsfIOmvkpZL+q2kTqWp3MyssFLtCU8AHoqIgcBD2XxDJgGXNNB+M/CDiDgKeAW4rChVmpkVWaoQvgC4I5u+AxjdUKeIeAh4Pb9NkoAzgbt3t76ZWWuXKoQPiYi12fQ64JAWrNsd2BQR27P51YAv5Ga2l5s2bRqSeOaZZ+rann32Wc455xwGDhzICSecwMUXX0xNTU3CKguvaCEsaaakxQ3cLsjvFxEBRBHrGCepSlLV+vXri7UZM3uXpkyZwqmnnsqUKVMAeOuttzj33HP5whe+wLJly3jyySf54he/SFv7Oy7a1ZYj4qzGlkmqkdQrItZK6gW81IK73gB0k9Qh2xs+DKhuoo5bgVsBysvLixb2ZrbnNm/ezGOPPcbs2bM5//zz+da3vsWdd97J8OHDOf/88+v6nX766emKLJJUwxHTgYpsugK4t7krZnvOs4GL9mR9M2t97r33Xs4++2wGDRpE9+7dmT9/PosXL2bYsGGpSyu6VCF8E/ARScuAs7J5JJVL+mVtJ0mPAr8DRkhaLWlktujrwFclLSc3RnxbSas3s4KaMmUKY8eOBWDs2LF1QxLtQdGGI5oSERuAEQ20VwGfy5v/UCPrrwROKlqBZlZ0056qZtKMpaxa+xLVf5rJ4/MXsN++HdixYweSuO6663jkkUdSl1l0/sacmZXctKequeaeRVRv2sIbS+ey3zFn0P1zv2Ty1LmsWrWKAQMGcNRRR/HnP/+Z+++/v269OXPmsHjx4oSVF55D2MxKbtKMpWzZtgOAN/7+CPsNGs6WbTuYNGMpABdeeCGVlZXcd999/OhHP2LgwIEcc8wx/OQnP6Fnz54pSy+4JMMRZta+rdm0pW760E99d5f2K664oq7tgQceKF1hCXhP2MxKrne3Li1qb8scwmZWclePPJouHct2auvSsYyrRx6dqKJ0PBxhZiU3+vjcmQYmzVjKmk1b6N2tC1ePPLquvT1xCJtZEqOP79MuQ7c+D0eYmSXkPWEz26tt2LCBESNy3/1at24dZWVldR9jmzFjBv/2b//GvHnzOOigg+jUqRNf+9rXGDNmTMqSd+IQNrO9Wvfu3VmwYAEAEydOpGvXrlx11VVEBKeccgoVFRXceeedALzwwgtMnz49Zbm78HCEmbVJs2bNolOnTowfP76urV+/fnz5y19OWNWuHMJm1iYtWbKEE044IXUZu+UQNrN24fLLL2fo0KGceOKJqUvZiceEzWyvVHsWtvzPGecbPHgwv//97+vmb7nlFl5++WXKy8tLXWqTvCdsZnud/LOwBVC9aQvX3LOIZ9a+VtfnzDPP5K233uKnP/1pXdubb76ZoNqmOYTNbK+Tfxa2Wlu27WDuig1185KYNm0ajzzyCAMGDOCkk06ioqKCm2++udTlNsnDEWa218k/C1u+DuUXc9VV59bN9+rVi8rKylKVtUe8J2xme522dBY2h7CZ7XXa0lnYPBxhZnudtnQWNoewme2V2spZ2JIMR0g6WNKDkpZlPw9qpN8DkjZJuq9e++2SnpO0ILsdV5rKzcwKK9WY8ATgoYgYCDyUzTdkEnBJI8uujojjstuCYhRpZlZsqUL4AuCObPoOYHRDnSLiIeD1UhVlZlZqqUL4kIhYm02vAw7Zg/v4tqSFkn4gad8C1mZmVjJFOzAnaSZwaAOLvpE/ExEhKVp499eQC+9OwK3A14HrG6ljHDAOoG/fvi3cjJlZcRUthCPirMaWSaqR1Csi1krqBbzUwvuu3Yt+W9J/AVc10fdWckFNeXl5S8PezKyoUg1HTAcqsukK4N6WrJwFN5JEbjx5cUGrMzMrkVQhfBPwEUnLgLOyeSSVS/plbSdJjwK/A0ZIWi1pZLboN5IWAYuAHsCNJa3ezKxAknxZIyI2ACMaaK8CPpc3/6FG1j+zeNWZmZWOzx1hZpaQQ9jMLCGHsJlZQg5hM7OEHMJmZgkpov18f0HSeuCFEm6yB/ByCbfXlNZUC7ie3XE9TWtt9RwdEQfsyYrt6nzCEdGzlNuTVBURreL62q2pFnA9u+N6mtYa69nTdT0cYWaWkEPYzCwhh3Bx3Zq6gDytqRZwPbvjeprWZuppVwfmzMxaG+8Jm5kl5BAuoOZcwFRSP0lPZhcoXSJpfMJajpP0l6yOhZI+WYxamltP1q/Bi7sWsI6zJS2VtFzSLtc2lLSvpN9my/8qqX8x6mhBPadlr5ftki4qZi3NrOerkv6WvV4ektQvcT3jJS3K/p4ek3RMynry+l0oKSTt/hMcEeFbgW7AfwATsukJwM0N9OkE7JtNdwWeB3onqmUQMDCb7g2sBbqlem6yZSOA84H7ilBDGbACOCL7PTwNHFOvzxeBn2XTY4HfFvH10px6+gPvB34NXFSsWlpQzxnAftn0F1rB8/OevOlRwAMp68n6HQDMAeYB5bu7X+8JF9ZuL2AaEVsj4u1sdl+K926kObU8GxHLsuk15K5wUqzPUreGi7ueBCyPiJURsRWozOpqrM67yZ3LWqnqiYjnI2Ih8I8i1dDSemZHxJvZ7DzgsMT1vJY3uz9QzINczXn9ANwA3Ay81Zw7dQgXVrMuYCrpcEkLgVXk9gjXpKolr6aTyP13X1GEWlpcT5H0Ifec11qdtTXYJyK2A68C3RPWU0otrecy4I+p65F0uaQV5N5tXZGyHkknAIdHxP3NvdN29Y25QijEBUwjYhXwfkm9gWmS7o6ImhS1ZPfTC/hvoCIi9niPq8gXd7VWRNI/A+XAh1PXEhG3ALdI+jRwLe9cOq2kJO0DfB+4tCXrOYRbKAp4AdOIWCNpMfAhcm99S16LpPcA9wPfiIh5La2h0PUUWTVweN78YVlbQ31WS+oAHAhsSFhPKTWrHklnkfvH+uG8obVk9eSpBH6asJ4DgCHAw9kI1qHAdEmjInfVoAZ5OKKwdnsBU0mHSeqSTR8EnAosTVRLJ2Aq8OuIaPE/gULXUwJPAAMlDcge+9isrnz5dV4EzIrsaEuiekppt/VIOh74OTAqIor9j7Q59QzMmz0XWJaqnoh4NSJ6RET/iOhPbsy8yQCuXdG3wh097Q48RO6FMBM4OGsvB36ZTX8EWEjuyOpCYFzCWv4Z2AYsyLsdl6qebP5RYD2whdyY28gC13EO8Cy5se9vZG3XZ38sAJ3JXVx2OfA4cESRXzO7q+fE7Hl4g9we+ZLE9cwEavJeL9MT1/NDYElWy2xgcMp66vV9mGZ8OsLfmDMzS8jDEWZmCTmEzcwScgibmSXkEDYzS8ghbGaWkEPY2hVJO7Izbi2W9DtJ+2Xth0qqlLRC0nxJf5A0qIH1fyXppexLNmbvmkPY2pstEXFcRAwBtgLjsxP0TAUejogjI2IYcA0Nn9/iduDsklVrbZ6/tmzt2aPkThN5BrAtIn5WuyAinm5ohYiYU+xzDFv74j1ha5ey80J8DFhE7vv+89NWZO2VQ9jamy6SFgBVwIvAbYnrsXbOwxHW3myJiOPyGyQtIXeyHrOS856wGcwC9pU0rrZB0vslfShhTdZOOISt3YvcWazGAGdlH1FbAnyX3BVAdiJpCvAX4GhJqyVdVtpqra3xWdTMzBLynrCZWUIOYTOzhBzCZmYJOYTNzBJyCJuZJeQQNjNLyCFsZpaQQ9jMLKH/DxbqHbOHlC1GAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 360x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# pca - code: https://towardsdatascience.com/pca-using-python-scikit-learn-e653f8989e60 \n",
    "from sklearn.decomposition import PCA\n",
    "\n",
    "X = weights\n",
    "\n",
    "pca = PCA(n_components=2)\n",
    "\n",
    "principalComponents = pca.fit_transform(X)\n",
    "\n",
    "principalDf = pd.DataFrame(data = principalComponents\n",
    "             , columns = ['PC 1', 'PC 2'])\n",
    "\n",
    "# display PCA for positions -- do certain positions cluster together \n",
    "# NOTE: this is embedded space (like word2vec for base positions)\n",
    "plt.scatter(principalDf['PC 1'], principalDf['PC 2'])\n",
    "labels = [reverse_map[letter] for letter in new_alph]#[str(pos_i) for pos_i in range(0,seq_len)]\n",
    "\n",
    "for i, label in enumerate(labels): \n",
    "    plt.annotate(label, (principalDf['PC 1'][i], principalDf['PC 2'][i]))\n",
    "plt.rcParams[\"figure.figsize\"] = 5,1\n",
    "plt.xlabel('PC 1')\n",
    "plt.ylabel('PC 2')\n",
    "plt.title('PCA over Learned Embedding Weights')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# look at saliency map for positive vs. negative classes\n",
    "# modified code from keras-vis package github page \n",
    "#https://github.com/raghakot/keras-vis/blob/master/examples/mnist/attention.ipynb\n",
    "from vis.visualization import visualize_saliency\n",
    "from vis.utils import utils\n",
    "from keras import activations\n",
    "\n",
    "class_idx = 1 # true off-target case \n",
    "y_test = test_bins\n",
    "indices = np.where(y_test[:, class_idx] == 1.)[0] # cases where class is positive \n",
    "\n",
    "# pick some random input from here.\n",
    "idx = indices[0]\n",
    "\n",
    "# Utility to search for layer index by name. \n",
    "# Alternatively we can specify this as -1 since it corresponds to the last layer.\n",
    "layer_idx = utils.find_layer_idx(model, 'preds')\n",
    "\n",
    "# Swap softmax with linear\n",
    "model.layers[layer_idx].activation = activations.linear\n",
    "model = utils.apply_modifications(model)\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np; np.random.seed(1)\n",
    "\n",
    "def viz_saliency_bar(model, data_idx, layer_idx=-1, class_idx=1, negate=False, plotting = True):\n",
    "    # assumes using test_data\n",
    "    # NOTE: b/c of embedding layer, this just gives gradients over positions not the actual numerical letters\n",
    "    if not negate: \n",
    "        grads = visualize_saliency(model, layer_idx, filter_indices=class_idx, seed_input=test_data[data_idx],\n",
    "                               wrt_tensor=model.layers[1].output)\n",
    "    else: \n",
    "        grads = visualize_saliency(model, layer_idx, filter_indices=class_idx, seed_input=test_data[data_idx],\n",
    "                                grad_modifier='negate', \n",
    "                               wrt_tensor=model.layers[1].output)\n",
    "    # diff colors in plot represent the saliency of specific positions \n",
    "    \n",
    "    # 1d heatmap code: https://stackoverflow.com/questions/45841786/creating-a-1d-heat-map-from-a-line-graph\n",
    "    # commented out code below will also give line graph \n",
    "    if plotting:\n",
    "        plt.rcParams[\"figure.figsize\"] = 5,1\n",
    "\n",
    "        x = list(range(0,23))\n",
    "        y = grads#np.cumsum(np.random.randn(50))+6\n",
    "\n",
    "        fig, ax = plt.subplots(ncols=1, sharex=True)\n",
    "\n",
    "        extent = [x[0]-(x[1]-x[0])/2., x[-1]+(x[1]-x[0])/2.,0,1]\n",
    "        # darker blue = larger saliency \n",
    "        ax.imshow(y[np.newaxis,:], cmap=\"jet\", aspect=\"auto\", extent=extent)\n",
    "        ax.set_yticks([])\n",
    "        ax.set_xlim(extent[0], extent[1])\n",
    "\n",
    "        plt.tight_layout()\n",
    "        plt.show()\n",
    "    return grads"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x72 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "grad = viz_saliency_bar(model, idx,plotting=True, negate=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_show = 20 # len(indices)\n",
    "grads = [] \n",
    "for idx in indices[:num_show]:\n",
    "    grad = viz_saliency_bar(model, idx,plotting=False,negate=False)\n",
    "    grads.append(grad)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(114.0, 0.5, 'Samples')"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# heatmap over all positive examples' (in test set's) saliency \n",
    "plt.rcParams[\"figure.figsize\"] = 15,5\n",
    "sns.heatmap(grads, cmap='jet')\n",
    "plt.xlabel('Position')\n",
    "plt.ylabel('Samples')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# average over samples and now plot the original alphabet characters on the y axis, colored by gradient \n",
    "# total saliency of each nt per position (then maybe avg)\n",
    "num_show = 20 # len(indices)\n",
    "grads_over_letters = {c: np.zeros(seq_len) for c in new_alph}\n",
    "counts_per_positon = {c: np.zeros(seq_len) for c in new_alph} # so we can normalize by counts at each position\n",
    "for idx in indices[:num_show]:\n",
    "    grad = viz_saliency_bar(model, idx,plotting=False,negate=False)\n",
    "    seq = test_df['Encoded Seqs'][idx]\n",
    "    for i, grad_i in enumerate(grad): \n",
    "        grads_over_letters[seq[i]][i] += grad_i \n",
    "        counts_per_positon[seq[i]][i] += 1 "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(114.0, 0.5, 'Samples')"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# heatmap over all positive examples' (in test set's) saliency \n",
    "plt.rcParams[\"figure.figsize\"] = 15,5\n",
    "sns.heatmap([grads_over_letters[letter] for letter in new_alph], cmap='jet',yticklabels=[reverse_map[letter] for letter in new_alph])\n",
    "plt.xlabel('Position')\n",
    "plt.ylabel('Samples')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kcollins/anaconda/envs/deeptoehold/lib/python3.6/site-packages/ipykernel_launcher.py:3: RuntimeWarning: invalid value encountered in true_divide\n",
      "  This is separate from the ipykernel package so we can avoid doing imports until\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Text(114.0, 0.5, 'Samples')"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# heatmap over all positive examples' (in test set's) saliency \n",
    "plt.rcParams[\"figure.figsize\"] = 15,5\n",
    "sns.heatmap([grads_over_letters[letter]/counts_per_positon[letter] for letter in new_alph], cmap='jet',yticklabels=[reverse_map[letter] for letter in new_alph])\n",
    "plt.xlabel('Position')\n",
    "plt.ylabel('Samples')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# look at saliency for true positives and false negatives\n",
    "grads_tp = []\n",
    "grads_fn = [] \n",
    "for sample_idx, (true_class, pred_class) in enumerate(zip(y_true_labels, y_pred_labels)): \n",
    "    if true_class == 1 and pred_class == 1: \n",
    "        grad = viz_saliency_bar(model, data_idx=sample_idx,plotting=False)\n",
    "        grads_tp.append(grad)\n",
    "    elif true_class == 1 and pred_class == 0: \n",
    "        grad = viz_saliency_bar(model, data_idx=sample_idx, class_idx=0,plotting=False)\n",
    "        grads_fn.append(grad)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1.0, 'Saliency Map: True Positives')"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# compare saliency heatmaps for true positive examples and cases where model thought incorreclty that sequence was negative \n",
    " \n",
    "plt.rcParams[\"figure.figsize\"] = 15,5\n",
    "sns.heatmap(grads_tp, cmap='jet')\n",
    "plt.xlabel('Position')\n",
    "plt.ylabel('Samples')\n",
    "plt.title('Saliency Map: True Positives')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1.0, 'Saliency Map: False Negatives')"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.rcParams[\"figure.figsize\"] = 15,5\n",
    "sns.heatmap(grads_fn, cmap='jet')\n",
    "plt.xlabel('Position')\n",
    "plt.ylabel('Samples')\n",
    "plt.title('Saliency Map: False Negatives')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "# average over samples and now plot the original alphabet characters on the y axis, colored by gradient \n",
    "# total saliency of each nt per position (then maybe avg)\n",
    "grads_over_letters_tp = {c: np.zeros(seq_len) for c in new_alph}\n",
    "counts_per_positon_tp = {c: np.zeros(seq_len) for c in new_alph} # so we can normalize by counts at each position\n",
    "grads_over_letters_fn = {c: np.zeros(seq_len) for c in new_alph}\n",
    "counts_per_positon_fn = {c: np.zeros(seq_len) for c in new_alph} # so we can normalize by counts at each position\n",
    "for sample_idx, (true_class, pred_class) in enumerate(zip(y_true_labels, y_pred_labels)): \n",
    "    if true_class == 1 and pred_class == 1: \n",
    "        grad = viz_saliency_bar(model, sample_idx,plotting=False,negate=False)\n",
    "        seq = test_df['Encoded Seqs'][sample_idx]\n",
    "        for i, grad_i in enumerate(grad): \n",
    "            grads_over_letters_tp[seq[i]][i] += grad_i \n",
    "            counts_per_positon_tp[seq[i]][i] += 1 \n",
    "    elif true_class == 1 and pred_class == 0: \n",
    "        grad = viz_saliency_bar(model, sample_idx,class_idx=0,plotting=False,negate=False)\n",
    "        seq = test_df['Encoded Seqs'][sample_idx]\n",
    "        for i, grad_i in enumerate(grad): \n",
    "            grads_over_letters_fn[seq[i]][i] += grad_i \n",
    "            counts_per_positon_fn[seq[i]][i] += 1 "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kcollins/anaconda/envs/deeptoehold/lib/python3.6/site-packages/ipykernel_launcher.py:4: RuntimeWarning: invalid value encountered in true_divide\n",
      "  after removing the cwd from sys.path.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1.0, 'Saliency Map: True Positives')"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAxkAAAFNCAYAAABsY6I3AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvOIA7rQAAIABJREFUeJzt3Xm4JGV99vHvzQwgICibKEvEDRU3RBSCxCW4QVAUFcF9ncQ3JIJGEfVFRGNcotE3ISajgivuSIhgxCiKaEQG2REUEWVTREAEFJjh9/7RdUJzPOdMn0NX9fTp7+e66pqupet++syZmvr1U09VqgpJkiRJGpa1Rt0ASZIkSYuLRYYkSZKkobLIkCRJkjRUFhmSJEmShsoiQ5IkSdJQWWRIkiRJGiqLDEljKcklSZ7UvH5Tko+Muk1qT5KXJPnqHOufkOS8LtskSZqdRYakkUmyW5LvJfltkmuSfDfJo+e7n6p6Z1W9so02LkSSjyWpJHtPW/5PzfKXtpz/piQ3NNMfkqzqm2/9RDzJ/ZvPOZX5sySvvzP7rKqPV9Uezf6XNvvftm/9t6rqIXeu5ZKkYbHIkDQSSTYCvgL8M7AJsBXwNuDmUbZriH4MvHhqJslSYF/gp20HN0XXXavqrsBfAf8zNT/TiXjTtjbaMdWGFwGHT/U8SZIWP4sMSaOyHUBVfaaqVlXV76vqxKo6GyDJ/ZJ8M8lvklyd5NNJ7j7TjpIcluRTffO7ND0k1yU5K8kT+tZ9K8nbm16T3yU5Mclmfet363vvpUlemuTRSX6VZEnfdvskOWuOz/efwG5JNm7mnwacDfyybx9zfsbmkrBDkpyf5NokRyW5y2A/3tn19QT8nyQXARdM9T5M2+6U/l6XJK9MckHTlq8m2WaQvKo6BbgAeGizn92SrGh6sH6QZOe+jFc0n/t3SS5Osl9f9reazU5u/jyv6Sl5dpInJbmk2fbNST477bMckeT9zeu7Nz/LK5NcluTwJGs167ZLcnLTtquTHD3YT1WS1M8iQ9Ko/BhYleTjSfboOxmfEuAfgC2BBwPbAIetbqdJtgKOB95Br4fk74AvJdm8b7PnAy8D7gGs02xDknsDX6XXu7I5sANwZlWdBvwGeErfPl4EfGKOpvwB+A9gv2b+xTNsP8hnfAHwVOB+9Aqzt/R91uuS7DZHG1bnGcCjgYetbsMkzwZeD+xN72dzKrDaE/D0/Bm9z3dGU9AdD7wP2JTez/qEJBs3vVvvB55cVRsCj6VXmE33uObPhzS9JV+atv4zwF5JNmjasBR4bl97Pwn8nt7P9FHAX9D7fQD4+6Z9GwNbA0es7jNKkv6YRYakkaiq64HdgAI+DPw6yXFJtmjWX1RVX6+qm6vq1/ROPh8/wK5fCJxQVSdU1W1V9XVgBbBn3zZHVdWPq+r3wOfpFRPQKz7+u+ldubWqflNVZzbrPt7smySb0DvxX91J9ieAFze9E48Hjp32MxjkM/5LVV1aVdfQOwHev+/9d296CRbqnVV1bfNzWJ2/ara/sKpW0iviHtMUdTNKch1wDbAceF1VfRt4OnBe8zNeWVWfBC6md6IPvd+Hhya5S1VdWVXnz/dDVdXFwLn0CiKAJwPXVtWKpr1PAg6qqpuq6lfAB7i9GLwV2Ba4V1X9oaq+O998SZJFhqQRqqofVdVLq2prepfSbEnvhI8kWyT5bJLLk1wPfArYbI7dTbk38NzmW/7rmhPd3YB79W3zy77XNwF3bV5vw+xjJj4FPL35dnxf4DtVdeVqPt8p9L71fzPwlekn8wN+xkv7Xv+c3s9oWC5d/Sb/697AEX0/06uB2+h92z+jpgjauKoeXFVTPQJb0vsc/X4ObNUUnvsDfw38MslXkmw3jzb2O5rbC7Lnc3tBeG9gXeBXfZ/lCGCLZv3rgLWBFUnOSfKSBeZL0kSzyJC0RqiqC4CP0Vy3D7yT3rfaD6uqjej1ImSAXV0KfLI5wZ2aNqiqdw343vvN0r7Lgf8B9qF3qdQnB9gf9AqH1zHzpVWDfMb+cQ9/AlwxYO4g+sdg3AiQZP2+Zffse30p8IppP9f1qurUeWZeQe9Ev9+fAJcDVNVXq+pJ9IrCi4B/X027Z/N54ElNz8Xe3F5kXEqvsNyk73NsVFUPb/KvrKpXVtW96BU7y5PcZ34fUZJkkSFpJJI8KMnrkmzdzG9D75vn7zebbAjcAPy2OVEc9BaoUz0OT02yJMld0nuGwqzfuPf5NL0T032bwdGbJtmhb/0ngDfQG8NwzIDt+X/0Ltc5eYZ1g3zGv06ydXOJ1puBzw2YO1+/bKYXNj+3ZdyxGPg34M1JHgz/O3j6OQvI+QrwkCTPa37GzwfuDxyf5F5Jnt4UOrfQK3xum76DqlpFb4zMfWcLqapfAqfQK1wvrKqfNMsvBb4N/GOSjZKs1Qx6f1zzufbtuwTsOnoFzaoFfE5JmmgWGZJG5XfAzsCpSW6kV1ycS+9bf+jdznZH4Lf0BuIOdFLfnETuDbwJ+DW9b65fzwDHu6r6Bb2xG6+jN5bgTOARfZt8md6J95er6qYB23NNVX2jqmb69n2Qz3g0cCK9cQs/pTcWAoDmzkp/Nkg7BmhnAa+i93O7mt6J/6l9679Ab8zIF5pLu86mNy5lvjm/pjfg/GB6hcJBwF5VdS2whN7f1ZXNul3p9SbM5K3A0c0lT/vMss3R9MZfTB8780JgA+B84FrgC9zea7MzcFrzO3kM8NfN74UkaR4y8/97kqSZJPkp8JdV9d8dZF0CvLKLLEmShsmeDEkaUHMb1wK+Oeq2SJK0JrPIkKQBNA+C+xC9y2f+aJyAJEnjKMmRSa5Kcu4s65Pk/yW5KMnZSXYcaL9eLiVJkiRNpubGFzcAn6iqh86wfk/gb+iNWdwZ+GBV7by6/dqTIUmSJE2oqjqZ3s1OZrM3vQKkqur7wN2T3GuO7QGLDEmSJEmz24o7Prz1smbZnJa21pw7KflY59dxrfOb2e6C2J5nbvLlzjPvznWdZx5x/YGdZ6592QguBXzI71e/zbB9cL3uM4H3/e1sdxZtz0d4ZeeZ539soEtPh2uu75Na8sDXntV55sG8u/PMR3JG55lXDPUh7YPb85pvdJ654yb/03nmmW/apfPMo/5hkOeCDtfLRvDvBaDqDSPJHYHu/1IX4G3JvE9uDoO/BJb1LVpeVcuH1qhZrLFFhiRJkqQ7pyko7kxRcTmwTd/81s2yOXm5lCRJkjQGli5gGoLjgBc3d5naBfhtVV05SFslSZIkreHWbmGfST4DPAHYLMllwFunoqrq34AT6N1Z6iLgJuBlg+zXIkOSJEkaA22cuFfV/qtZX8C8B2JaZEiSJEljoI2ejLZYZEiSJEljYJxO3MeprZIkSdLEsidDkiRJ0lCN04n7OLVVkiRJmlj2ZEiSJEkaqnE6cR+ntkqSJEkTa5x6Mjp74neSbZK8vqs8SZIkaTFZewHTqLTak5Fkc+C5wP7AlsCX28yTJEmSFqtxugRp6G1NsiGwD/B8YDvgGOA+VbX1AO9dBizrzb2E3hPOJUmSJI3T5VJtFERXAT8A3gKcUlWV5FmDvLGqlgPLAZKPVQttkyRJksbSOPVktDEm4xBgXeBfgUOS3K+FDEmSJGmijNOYjKEXGVX1garaBdi7WXQssGWSg5NsN+w8SZIkaRIsXcA0Kq3dXaqqLq6qd1bVw4CdgKcAX20rT5IkSdKaoe27Sz2S3p2l9gUuAT7QZp4kSZK0WE30wO/mkqj9m+lq4HNAquoJw86SJEmSJsU4Dfxuo60XAN8B9qqqiwCSHNRCjiRJkjQxxqkno40xGfsAVwInJflwkt2BtJAjSZIkTYxxGvg99OyqOhY4NskG9O4wdSBwjyQfAr5cVScOO1OSJEla7Ca9JwOAqrqxqo6uqqcDWwNnAAe3lSdJkiQtZuP0nIxOelGq6lp6T/Je3kWeJEmStNhM+sBvSZIkSUO29hiduaeqRt2GGSUf675hT39p55EA+x738U7zvsUTO82bcvn19+40b+1jR/C7fUn3kS869MPdhzZ24MxO8w655h86zQO4+T136zzzQ+/uPBKAD9RZneadd/0jOs0DuGijbo9DAA/67s87z5xSD+42756bdP9ZV922pPPMqx+wdeeZuXg0B4aqN4wkdwTG4iZF16+XeZ/cbPT7GslnW2ProaqXdp6ZZ3QeCcDneEmneVvwi07zpizdaM0saLVwr+WITvMO4fpO8ybNhTy807yVnaZNpmzS9XF3NP+/SJNi7e5r6gVbY4sMSZIkSbdbOkZn7mPUVEmSJGlyjdOYjDFqqiRJkjTBvFxKkiRJ0lCN0Zn7GDVVkiRJmmBjdOY+Rk2VJEmSJtgYnbmvNeoGSJIkSVpcxqgekiRJkiaYA78lSZIkDdUYnbmPUVMlSZKkCTZGZ+5DH5OR5MRh71OSJEmaeEsWMI1IG/XQ5i3sU5IkSZpsY9ST0UZT75Zkn9lWVtUxs61LsgxYBvDv//7vLFu2rIXmSZIkSWNo0osMYC8gM6wrYNYio6qWA8v7tpUkSZIEE393qZ9X1ctb2K8kSZI0ucaoJ6ONh/HN1INBkt2SHNFCniRJkrT4LV3ANCJtRL9o6kWSRwLPB54L/Iw5LpWSJEmSNIcx6sloo6m3JHkrsD9wNfA5IFX1xBayJEmSpMkw4WMyLgC+A+xVVRcBJDmohRxJkiRpcoxRT0YbYzL2Aa4ETkry4SS7M8s4DUmSJEkDGqMxGUMvMqrq2KraD3gQcBJwIHCPJB9K8pRh50mSJEkTYYye+N1GTwYAVXVjVR1dVU8HtgbOAA5uK0+SJEla1Ca5J2MmVXVtVS2vqt27yJMkSZIWnZaKjCRPS3JhkouSvHGG9X+S5KQkZyQ5O8meq9tnJ0WGJEmSpDVPkiXAEcAewPbA/km2n7bZW4DPV9Ujgf2Af13dfi0yJEmSpHHQzpiMxwAXVdXFVXUL8Flg72nbFLBR8/puwBWr2+kY3QirA9d1H7neV65l/eu7zdxho592Gwh8lv34RceZ93jxaR0n9lx1+J90mnc1m3aaN+WRnMmbObTTzN032aHTPAA26D7y1YfDh7r90U6M69i488zvP3ZjdvnumZ3nnvbYhwIP6zTzYB7XaR4Aa8E7b3tTp5H/eXGncY1bRxGqNc0CztyTLAOW9S1aXlXL++a3Ai7tm78M2Hnabg4DTkzyN/T+Z3xSC02VNJfq+ORwtRdFaiy9umrUTWjd0o1G8RkfOYJM4LGj+KzdFhijdPVaW3Wa95+dpkl9FnDm3hQUy1e74dz2Bz5WVe9L8qfAJ5M8tKpum+0NFhmSJEnSOGjnzP1yYJu++a2bZf1eATwNoKr+J8ldgM2Aq2bbqWMyJEmSpHHQzpiM04AHJLlPknXoDew+bto2vwB2B0jyYOAuwK/n2qk9GZIkSdI4aOHMvapWJjkA+Bq9suTIqjovyeHAiqo6Dngd8OEkB9EbBP7Sqrmv67XIkCRJksZBS2fuVXUCcMK0ZYf2vT4feOx89mmRIUmSJI2DwS5/WiNYZEiSJEnjYIzO3MeoqZIkSdIEG6Mz9zFqqiRJkjTBxujMfYyaKkmSJE0wx2RIkiRJGqoxOnMfelOTbA5s3tzqqn/59sCvq2rOB3dIkiRJmsEYFRltPPH7n+k9Zny6TYEPtpAnSZIkLX7tPPG7FW0UGfevqpOnL6yq7wAPn+uNSZYlWZFkxfLly1tomiRJkjSmli5gGpE2ojecY93ac72xqpYDU9XFnI8qlyRJkrRmaqMn46Ike05fmGQP4OIW8iRJkqTFb8J7Mg4Ejk+yL3B6s2wn4E+BvVrIkyRJkha/MbqF7dB7MqrqJ8DDgG8D2wL3Br4FvBJ4zbDzJEmSpIkw4T0ZVNXNwFFJdgT2B94K/Az4Uht5kiRJ0qI3RrewbeM5GdvRKyz2B64GPgekqp447CxJkiRpYkxykQFcAHwH2KuqLgJIclALOZIkSdLkmOQxGcA+wJXASUk+nGR3IC3kSJIkSZNjjMZktDHw+9iq2g94EHASvbtN3SPJh5I8Zdh5kiRJ0kSY5CJjSlXdWFVHV9XTga2BM4CD28qTJEmSFrUlC5hGpJP6pqqupfck7+Wr21aSJEnSDCZ84LckSZKkYRujM/c1tqk5svvMe5z8i84zt+SKzjN/yv06z/wEL+o886ab1u8884Z3bTaCuxzs07sgsWO7Lvte55m/YbPOM3f+v9/m1Lc/vtPM8/7v/YD7d5p5FC/tNA/gzbyz88zN2LTzzCvYkpPZufPc+3Fl55mj+P/lX5/7Wl7XcWaN4tHCH9yo88gT+Fu+mrd0mrmsftJp3pRLOz7mLtgY3V1qjS0yJA1oBAXGxPm/1XHgmPxnJ0nq1hiduY9RUyVJkqQJNkZn7mPUVEmSJGmCjdGZ+xg1VZIkSZpgjsmQJEmSNFRjdObe2sP4JEmSJE2mMaqHJEmSpAk2RmfuY9RUSZIkaYI5JkOSJEnSUI3RmfsYNVWSJEmaYGN05r7apib5fzMs/i2woqr+Y/hNkiRJkvRHxqjIGOTuUncBdgB+0kwPB7YGXpHkAy22TZIkSVKjlsx/GpVB6qGHA4+tqlUAST4EfAfYDThn+sZJdgPuW1WfaOa/CGzSrH5HVX1zGA2XJEmSJsmqBfRkjKrzY5CejI2Bu/bNbwBs0hQdN8+w/duAFX3zDwReDxwGvGGuoCTLkqxIsoJvLx+gaZIkSdJkWLV0/tOoDBL9HuDMJN8CAjwOeGeSDYD/nmH7jarq/L75n1TV6QBJ/mGuoKpaDiwHyJHUAG2TJEmSJsLKJfN/jva6LbRjEKstMqrqo0lOAB7TLHpTVV3RvH79DG+5+7T379M3u8WCWilJkiRNuFVLx2fk96AtXQv4dbP9/ZPcv6pOnmXbC5L8RVUd378wyV7AhQtvqiRJkjS5Vi0Zn6fxDXIL23cDzwPOA25rFhcwW5FxEHB8kucAP2yWPQrYFdjrTrVWkiRJmlCrxuiR34P0ZDwTeGBVzTTI+49U1UVJHg68AHgIvYLk28CRwGuAv15gWyVJkqSJtXKRFRkXA2sz852kZtQUJEcm2RHYH3gr8DPgSwtppCRJkjTpVo3R0/gGaelN9O4u9Q36Co2q+tuZNk6yHb3CYn/gauBzQKrqiXe+uZIkSdJkWmyXSx3XTIO6gN7D+vaqqosAkhy0gLZJkiRJaiyqIqOqPj7Pfe4D7AeclOS/gM/Se76GJEmSpAkw6xM9kny++fOcJGdPn2Z7X1UdW1X7AQ8CTgIOBO6R5ENJnjLsDyBJkiRNglUsmfc0iCRPS3JhkouSvHGWbfZNcn6S85Icvbp9ztWT8ZrmzwXddraqbgSOBo5OsjHwXOBg4MSF7E+SJEmaZG3cXSrJEuAI4MnAZcBpSY6rqvP7tnkAcAjw2Kq6Nsk9VrffWYuMqrqyCf3YnR20XVXXAsubSZIkSdI8tXR3qccAF1XVxQBJPgvsDZzft82rgCOac3qq6qrV7XTOllbVqiS3JblbVf12wU1fgHWeeX2XcQB8ihd0nnkL63ae+YJbPtV55qp1ur/l2g3/uVnnmTyt+0iAJ+z6X51n7sTpnWdeyjadZ/77bgd2nrnWdzuPZKfq/u/zcA7tPPNdNx3SeebN63d/nAe4gnt1nnk2D+s8c7svzHoFd3t26z4Sft954p68mxM4uPNcza6lgd9bAZf2zV8G7Dxtm+0AknwXWAIcVlVznnwMcuZ3A3BOkq8DN04tnO0WtpK6NYoCQ5I0Gfao6jTv0tVvMtEWUmQkWQYs61u0vKrme3XRUuABwBOArYGTkzysqq6b6w2rc0wzSZIkSRqRhYzJaAqKuYqKy+EOlwps3SzrdxlwalXdCvwsyY/pFR2nzbbTNm5hK0mSJGnIWhqTcRrwgCT3oVdc7Ac8f9o2x9J70PZRSTajd/nUxXPtdNaWJvl8Ve2b5Bzgj/rKqurh82u/JEmSpIVqY0xGVa1McgDwNXrjLY6sqvOSHA6sqKrjmnVPSXI+sAp4fVX9Zq79tnYLW0mSJEnD09YTv6vqBOCEacsO7XtdwGubaSBz3sK2+fPn826pJEmSpKFqq8how6xP/J6SZJckpyW5IcktSVYl6f7+spIkSdIEW8mSeU+jMsjokX+hNwDkC8BOwItp7pUrSZIkqRstDfxuxWp7MgCq6iJgSVWtqqqjGNkjxyRJkqTJtIol855GZZBy6KYk6wBnJnkPcCUDFieSJEmShmNRjckAXtRsdwC9J35vAzx7to2TjE8/jiRJkjQmFs2YjCQ7APcHzquqHwFvG2CfPwB2HELbJEmSJDUWxZiMJIcCn6fXa3F8klcNuM8Mo2GSJEmSxtNc5dDzgB2q6qYkmwL/BXx4gH1unmTWB3VU1ftnW5dkGbAMYMn7P8CSl7xsgDhJkiRp8RunMRlzFRk3V9VNAFX1mySDDvZeAtyVBfRoVNVyYDnAutdcX/N9vyRJkrRYLZYi475JjmteB7hf3zxV9YxZ3ndNVR0+rAZKkiRJYqQDuedrriJj72nz/zjgPm9eYFskSZIkzWKcBn7P2tKq+vYC9zk+JZYkSZI0JhbL5VILtelCB35LkiRJmtmkFxkLHvgtSZIkaWaLsshIsv7U3aZW40oHfkuSJEnDNU4Dv1d7W9okuyY5H7igmX9Ekn+d6y3DapwkSZKknlUsnfc0KoMk/xPwVOA4gKo6K8nj5th+92E0TJIkSdLtFt3lUlV1aXKHDopVc2x7zZ1tlCRJkqQ7WmxFxqVJdgUqydrAa4AftdssSZIkSf3GaUzGIEXGXwEfBLYCLgdOBP66zUZJkiRJuqNF8TC+KVV1NfCCDtoycgdwROeZD+OczjO3X6f7jqgv8uzOM/la95F3+7dfdp55xi07sNM6p3ee+yvu0Xnmfbik88x3nXIgb9ztA51mfr7TtJ5ncGHnmWeyQ+eZb13/MN5202GdZj7kiz/tNG/Kd5/zqM4z/4RLO88855aHdZ7Jy7qP5Lu/H0Go1jSL4nKpJP8M1Gzrq+pvW2lR4+ZNNmpz9zN6YOeJ0p03igJj4pwy66GwHfEmfW1ad/2O/z696aKkIRmnImOuW9iuAE4H7gLsCPykmXYA1mm/aZIkSZKmrGLJvKdRmbUno6o+DpDk1cBuVbWymf834DvdNE+SJEnSuBlk9MjGwEbA1K1p79oskyRJktSRxXZ3qXcBZyQ5id6FpY8DDmuzUZIkSZLuaLHdXeqoJF8Fdm4WHVxV3d9CR5IkSZpg4zTwe667Sz2oqi5IsmOzaOq+dFsm2bKqfth+8yRJkiTBIikygNcCy4D3zbCugD9vpUWSJEmS/siiGJNRVcual3tU1R/61yW5S6utkiRJknQH4zQmY67nZEz53oDLJEmSJLVkUTwnI8k9ga2A9ZI8ktsfWboRsH4HbZMkSZLUWCxjMp4KvBTYGnh/3/LfAW+a7U1J7g9sUVXfnbb8scAvq+qnC26tJEmSNKEWRZHRPPH740meXVVfmsc+PwAcMsPy65t1T59fEyVJkiSN08DvQcZkfCPJ+5OsaKb3JbnbHNtvUVXnTF/YLNt2rqAky6Zyli9fPkDTJEmSpMmwiqXznkZlkOSPAucC+zbzLwKOAvaZZfu7z7Gv9eYKqqrlwFR1UQO0TZIkSZoIi+JyqT73q6pn982/LcmZc2y/IsmrqurD/QuTvBI4fSGNlCRJkibdYisyfp9kt6o6Bf53APfv59j+DcDnkryA24uKnYB1gGfdmcZKkiRJk2qcxmQMUmS8mt4A8KlxGNfSu+vUbP6jqnZM8ufAQ5plx1fVNxfeTEmSJGmyjdPD+Fbb0qo6E3hEko2a+etX85Y0230TsLCQJEmShmBRXS6VZAvgncCWVbVHku2BP62qj87yls2TvHa2/VXV+2dbJ0mSJGn8DXIL248BXwO2bOZ/DBw4x/ZLgLsCG84ySZIkSZqnVSyZ9zQqg1zYtVlVfT7JIQBVtTLJqjm2v7KqDh9O8yRJkiTB4hv4fWOSTWmeW5FkF+C3c2yfYTRMkiRJ0u3GaeD3IJdLvRY4Drhfku8CnwD+Zo7tdx9GwyRJkiTdrq3LpZI8LcmFSS5K8sY5tnt2kkqy0+r2OcjdpX6Y5PHAA+n1UlxYVbfOsf01q9unJEmSpPlpY4xFkiXAEcCTgcuA05IcV1XnT9tuQ+A1wKmD7HfWIiPJPrOs2i4JVXXMQC2XJEmSdKe1NJD7McBFVXUxQJLPAnsD50/b7u3Au4HXD7LTuXoynj7HugIsMiRJkqSOtDTweyvg0r75y4Cd+zdIsiOwTVUdn+TOFRlV9bKFtHJY8vJRpD6c7Y48u9PEL/34BZ3mAXBJ95E8tfvI9X57beeZt/xhXda5y82dZm7JFZ3mTfkeu44k9z4j+QXu1lufO4LQfX/EGZ9/8AiCpTtn53V+0H3oKd1HDnavnuE6n7fwoxzWaeb2P72t07wpdd/xuG/RQgZ+J1kGLOtbtLyqls/j/WsB7wdeOp/cQR7G907gPVV1XTO/MfC6qnrLfILGxYU8vNO88fiV1nxct849O817cadpo/dG/mnUTVi0HvlHPeNt502I59RIYh87IZlP5bgRpEqjsZDLpZqCYq6i4nJgm775rZtlUzYEHgp8KwnAPYHjkjyjqlbMttNB7i61x1SB0TT0WmDPAd4nSZIkaUhaurvUacADktwnyTrAfnB79V5Vv62qzapq26raFvg+MGeBAYP1vS1Jsm5V3QyQZD1g3UFaLEmSJGk42hiT0Txo+wDga8AS4MiqOi/J4cCKqlpQd+EgRcangW8kOaqZfxnw8YWESZIkSVqYth7GV1UnACdMW3boLNs+YZB9DvKcjHcnOZvbH7L39qr62iA7lyRJkjQcLd3CthUDlUNV9VXgqy23RZIkSdIsFlWRkeR39J6LAbAOsDZwY1Vt1GbDJEmSJN1u1W0LKDIGuc1TCwa5XGrDqdfp3bdqb2CXNhslSZIk6Y5WrlxAkbHO8NsxiHnVNtVzLCN5tJokSZI0uVatXDrvaVQGuVxqn77ZtYCdgD+01iJJkiRJY22Q8ubpfa9XApfQu2RKkiRJUkdWLeRyqREZZEzGy+azwyQmapkyAAAVt0lEQVR7A1tX1RHN/KnA5s3qN1TVF+fdSkmSJGnCLYoiI8mMD+BoVFW9fZZ1b6D3OPIp6wKPBjYAjgIsMiRJkqR5Wnnr+BQZcw38vnGGCeAVwMFzvG+dqrq0b/6UqvpNVf2CXqExqyTLkqxIsoILl6++9ZIkSdKEuG3V0nlPozJrclW9b+p1kg2B1wAvAz4LvG+29wEbT9vPAX2zmzOHqloOLAfIy//32RySJEmSxuhyqTlvYZtkkyTvAM6mV5DsWFUHV9VVc7zt1CSvmmFffwn84E61VpIkSZpUK5fMfxqRucZkvBfYh17PwsOq6oYB93kQcGyS5wM/bJY9it7YjGfeibZKkiRJk2tlRt2Cgc11odbrgJuBtwBv7j3sG4DQG/i90Szv+6+q2jHJ7sD2zbLjq+qbw2iwJEmSNJFWjroBg5trTMa8ngbeJ837vwF8Y4H7kCRJktRvMRQZd8LmSV4728qqen8LmZIkSdLiNuFFxhLgrjQ9GpIkSZKG4NZRN2BwbRQZV1bV4S3sV5IkSZpcq0bdgMG1UWTYgyFJkiQN24RfLrV7C/uUJEmSJtskFxlVdc2w9ylJkiRNvEkuMiRJkiS1YIyKjFTVqNswoxxD5w07a58Hdh3JpWzTeebreW/nmW/m7zvPfOHRX+w8k227j+Seo/k3fMp9H9155r/yfzrP/PTzX9F5Jk/pPvKYl+7ReeaTVnX/KKWNTryl88yf7XGvzjMB3srbOs/clks6z/zH6/+u88x33G2TzjNfx792nglwfsfH3e1/eluneVPqvhmLMcU5Yf7nx7XnaMZL25MhSZKkGT244y+j18yvvtcgY9STYZEhSZIkjQOLDEmSJElDNeEP45MkSZI0bBP+MD5JkiRJw+blUpIkSZKGyiJDkiRJ0lBZZEiSJEkaKosMSZIkSUNlkSFJkiRpqCwyJEmSJA3VGD0nY602dppk8yTbz7B8+ySbt5EpSZIkLWqrFjCNSCtFBvDPwGYzLN8U+GBLmZIkSdLitXIB04i0VWTcv6pOnr6wqr4DPHy2NyVZlmRFkhWcuLylpkmSJEljaIyKjLbGZGw4x7q1Z1tRVcuB5QA5hhp2oyRJkqSxNUYDv9vqybgoyZ7TFybZA7i4pUxJkiRp8bp1AdOItNWTcSBwfJJ9gdObZTsBfwrs1VKmJEmSpDVAW0VGAS8HHgA8tFn2beDDjHScuyRJkjSmxugsuq0i4wPAIVV1VP/CJA9r1j29pVxJkiRpcRqjMRltFRlbVNU50xdW1TlJtm0pU5IkSVq8LDK4+xzr1mspU5IkSVq8Jv2J38CKJK+avjDJK7l9ILgkSZKkQY3RE7/bvLvUl5O8gDveXWod4FktZUqSJEmL16RfLlVVvwJ2TfJEbr+71PFV9c028iRJkqRFr6UiI8nTgA8CS4CPVNW7pq1/LfDKpgW/Bl5eVT+fa59t9WQAUFUnASe1mSFJkiRNhBbGZCRZAhwBPBm4DDgtyXFVdX7fZmcAO1XVTUleDbwHeN5c+21rTIYkSZKkYWpnTMZjgIuq6uKqugX4LLB3/wZVdVJV3dTMfh/YenU7bbUn4055YfeRj/iLCzvPfPAXzug8cxQO59DuQ+/SfSTvWv0mw/bM4z7XfSjwPXbtPPO4m57ReeaGH/k1v3vl5t2GntttHMA+f/dVjvnHPTrNXP/GWzrNA1j5WFj63c5jJ8YKHtV55hM2OolvXf/ETjNfu0+ncQC87pirug/VmmcBl0slWQYs61u0vKqW981vBVzaN38ZsPMcu3wF8NXV5a65RYYkrSmOrm7z/i7d5jX24YRO81Yyms/JHt3+fd6n07TbfWIEmXtyzAhS4aaNNh5JrtS5BRQZTUGxfLUbDiDJC+ndzOnxq9vWIkOSJEkaB+08J+NyYJu++a2bZXeQ5EnAm4HHV9XNq9upRYYkSZI0Dtp57sVpwAOS3IdecbEf8Pz+DZI8Evh34GlVNdC1exYZkiRJ0jho4Ra2VbUyyQHA1+jdwvbIqjovyeHAiqo6DngvcFfgC0kAflFVcw6UtMiQJEmSxkFLz8moqhPgjgPzqurQvtdPmu8+LTIkSZKkcdDOmIxW+JwMSZIkSUNlT4YkSZI0DtoZ+N0KiwxJkiRpHLQ0JqMNFhmSJEnSOLDIkCRJkjRUYzTw2yJDkiRJGgeTPiYjyW7AfavqE838F4FNmtXvqKpvtpErSZIkLVpjdLlUW7ewfRuwom/+gcDrgcOAN8z2piTLkqxIsoKVy1tqmiRJkjSGVi5gGpG2LpfaqKrO75v/SVWdDpDkH2Z7U1UtB5YDZH2qpbZJkiRJ48cxGdy9f6aq9umb3aKlTEmSJGnxGqMxGW1dLnVBkr+YvjDJXsCFLWVKkiRJi5eXS3EQcHyS5wA/bJY9CtgV2KulTEmSJGnxGqOB323ewvYVwHbAQ5r5k4GPMFYdPZIkSdIawjEZfAA4pKqO7F+Y5GHNuqe3lCtJkiQtTmP0VX1bRcYWVXXO9IVVdU6SbVvKlCRJkhavMbr3alsDv+8+x7r1WsqUJEmStAZoq8hYkeRV0xcmeSVwekuZkiRJktYAbV0udSDw5SQv4PaiYidgHeBZLWVKkiRJWgO0UmRU1a+AXZM8EXhos/j4qvpmG3mSJEmS1hxt3sKWqjoJOKnNDEmSJGkyLOQetmsPvRWDSNWaOUw97+5+/Px7Dv7briP5b57UeeYV3KvzzEtv2abzzN/fsH7nmbdctlHnmQCs6D7y7S//u84zf8T2nWd++tmv6DyTB3Uf+bK//1Dnme/gLZ1nrmJJ55mb3nx155kAX1v3qZ1nvoW/7zzzR6c+svPMelU6z8w57+488+sc3Hnmc26+svNMgOvWuWf3f6kLkPx+3ufHVeuN5LO12pMhqQMjKDAkSdIoLKQnYzQ3drXIkCRJksbCylE3YGAWGZIkSdJYWEhPxmhYZEiSJEljwSJDkiRJ0lB5uZQkSZKkobInQ5IkSdJQ2ZMhSZIkaajsyZAkSZI0VPZkSJIkSRoqezIkSZIkDZU9GZIkSZKGasJ7MpI8BLhfVR3XzP8TcLdm9b9U1Q/byJUkSZIWr/HpyVirpf2+C7i6b/6pwPHAScChs70pybIkK5Ks4NTlLTVNkiRJGke3LmAajbYul7pXVX2vb/76qvoSQJK/nO1NVbUcWA6Qd1MttU2SJElSi9oqMjbsn6mqXfpm79FSpiRJkrSIjc/lUm0VGVck2bmqTu1fmGQX4IqWMiVJkqRFbMIHfgMHA59L8jFgapD3o4CXAvu2lClJkiQtYhNeZFTVD5LsDBxAr7Ao4DzgMVV1VRuZkiRJ0uI24ZdLJdkb2LqqDm3mf0CvJ+NFSd5QVV9sI1eSJElavCa8JwN4A7Bf3/w69IqMuwJHARYZkiRJ0rxMeE8GsE5VXdo3f0pVXQNck2SDljIlSZKkRcyejI37Z6rqgL7ZzVvKlCRJkhax8enJaOuJ36cmedX0hc2D+H7QUqYkSZK0iLXzxO8kT0tyYZKLkrxxhvXrJvlcs/7UJNuubp9t9WQcBByb5Pnc8Ra26wLPbClTkiRJWsSG35ORZAlwBPBk4DLgtCTHVdX5fZu9Ari2qu6fZD/g3cDz5tpvW7ewvQrYNcmfAw9pFh9fVd9sI0+SJEla/FoZk/EY4KKquhggyWeBvYH+ImNv4LDm9ReBf0mSqqrZdtpWTwYATVFhYSFJkiTdaa2MydgK6L9h02XAzrNtU1Urk/wW2BS4eta9VtWim4BlZppp5pqba6aZZq65mZP0Wc00cxImYBmwom9aNm39c4CP9M2/CPiXaducS+8ZeFPzPwU2myu3rYHfo7bMTDPNXKNzzTTTzDU3c1S5Zpo5jplrvKpaXlU79U3Lp21yObBN3/zWzbIZt0myFLgb8Ju5chdrkSFJkiRp9U4DHpDkPknWofdA7eOmbXMc8JLm9XOAb1bTpTGbVsdkSJIkSVpzVW+MxQHA14AlwJFVdV6Sw4EVVXUc8FHgk0kuAq6hV4jMabEWGdO7gcw008w1K9dMM81cczNHlWummeOYuShU1QnACdOWHdr3+g/Ac+ezz6ymp0OSJEmS5sUxGZIkSZKGalEVGat7JHpLmUcmuSrJuV3kNZnbJDkpyflJzkvymg4y75LkB0nOajLf1nZmX/aSJGck+UpHeZckOSfJmUlWdJR59yRfTHJBkh8l+dOW8x7YfL6p6fokB7aZ2eQe1Pz+nJvkM0nu0kHma5q889r8jDMdC5JskuTrSX7S/LlxB5nPbT7rbUl2GmbeHJnvbX53z07y5SR37yDz7U3emUlOTLJl25l9616XpJJs1nZmksOSXN73b3XPtjOb5X/T/J2el+Q9bWcm+VzfZ7wkyZnDzJwjd4ck35863id5TAeZj0jyP83/M/+ZZKMh5s14ftDBsWi23NaOR3Nktno80jyM+t69Q7wH8BJ69+y9L7AOcBawfQe5jwN2BM7t8LPeC9ixeb0h8OO2PysQ4K7N67WBU4FdOvq8rwWOBr7SUd4lrObezy1kfhx4ZfN6HeDuHWYvAX4J3LvlnK2AnwHrNfOfB17acuZD6d3be316Y9D+G7h/S1l/dCwA3gO8sXn9RuDdHWQ+GHgg8C1gp44+51OApc3rd3f0OTfqe/23wL+1ndks34be4MifD/s4McvnPAz4u2H/Pa4m84nNv5V1m/l7dPGz7Vv/PuDQjj7ricAezes9gW91kHka8Pjm9cuBtw8xb8bzgw6ORbPltnY8miOz1eOR0+DTYurJ+N9HolfVLcDUI9FbVVUn0xtl35mqurKqfti8/h3wI3oncG1mVlXd0Myu3UytD+hJsjXwF8BH2s4alSR3o/cf0UcBquqWqrquwybsDvy0qn7eQdZSYL307rG9PnBFy3kPBk6tqpuqaiXwbWCfNoJmORbsTa+ApPnzmW1nVtWPqurCYeYMkHli8/MF+D69e6y3nXl93+wGDPl4NMex/Z+ANww7bzWZrZkl89XAu6rq5mabqzrIBCBJgH2Bzwwzc47cAqZ6Eu7GkI9Js2RuB5zcvP468Owh5s12ftD2sWjG3DaPR3Nktno80uAWU5Ex0yPRWz3xXhMk2RZ4JL2ehbazljRd2FcBX6+q1jOBD9D7D/22DrKmFHBiktOTdPFgn/sAvwaOSu+ysI8k2aCD3Cn70cJ/6NNV1eXAPwK/AK4EfltVJ7Ycey7wZ0k2TbI+vW8qt1nNe4Zpi6q6snn9S2CLDrNH5eXAV7sISvL3SS4FXgAcurrth5C3N3B5VZ3VdtY0BzSXfhw57MtcZrEdvX83pyb5dpJHd5A55c+AX1XVTzrKOxB4b/N79I/AIR1knsftX4I+l5aOSdPODzo7FnV5XjJAZmfHI/2xxVRkTJwkdwW+BBw47Vu9VlTVqqragd63Ao9J8tA285LsBVxVVae3mTOD3apqR2AP4K+TPK7lvKX0utM/VFWPBG6k153duvQeuvMM4AsdZG1M7z/W+wBbAhskeWGbmVX1I3rd5ScC/wWcCaxqM3OOthQd9P6NUpI3AyuBT3eRV1VvrqptmrwD2sxqitQ30UExM82HgPsBO9Arzt/XQeZSYBNgF+D1wOebHoYu7E8HX3r0eTVwUPN7dBBNj3LLXg78nySn07vM55ZhB8x1ftDmsajr85K5Mrs+HumPLaYiY5BHoi8aSdam94/q01V1TJfZzaU8JwFPaznqscAzklxC7/K3P0/yqZYzp75xn7pE4Mv0LsVr02XAZX09Q1+kV3R0YQ/gh1X1qw6yngT8rKp+XVW3AscAu7YdWlUfrapHVdXjgGvpXbfblV8luRdA8+dQLztZkyR5KbAX8ILmJKZLn2aIl5zM4n70CuSzmmPS1sAPk9yzzdCq+lXzBc9twIdp/3gEvWPSMc1lsj+g15M81EHuM2kuo9wH+FzbWX1eQu9YBL0vW1r/+VbVBVX1lKp6FL2C6qfD3P8s5wetH4tGcV4yW+aIj0dqLKYiY5BHoi8KzTdKHwV+VFXv7yhz86k7NCRZD3gycEGbmVV1SFVtXVXb0vv7/GZVtfrNd5INkmw49ZreALJW7xxWVb8ELk3ywGbR7sD5bWb26fJbw18AuyRZv/kd3p3eNbStSnKP5s8/oXcCc3TbmX2Oo3cSQ/Pnf3SY3ZkkT6N3WeMzquqmjjIf0De7N+0fj86pqntU1bbNMekyeoNOf9lm7tSJYeNZtHw8ahxLb/A3SbajdzOKqzvIfRJwQVVd1kHWlCuAxzev/xxo/TKtvmPSWsBbgH8b4r5nOz9o9Vg0ovOSGTNHcTzSLOYzSnxNn+hdb/1jet8KvLmjzM/Q68K+ld5/Oq/oIHM3el2dZ9O7/ONMYM+WMx8OnNFknksLd/5YTf4T6ODuUvTuTnZWM53X4e/RDsCK5ud7LLBxB5kbAL8B7tbh3+Pb6J0Mngt8kubuNS1nfode0XYWsHuLOX90LAA2Bb5B78Tlv4FNOsh8VvP6ZuBXwNc6yLyI3pi4qePRsO/0NFPml5rfo7OB/6Q34LPVzGnrL2H4d5ea6XN+Ejin+ZzHAffqIHMd4FPNz/eHwJ938bMFPgb81TCzBvisuwGnN8eHU4FHdZD5GnrnKj8G3kXzYOQh5c14ftDBsWi23NaOR3Nktno8chp88onfkiRJkoZqMV0uJUmSJGkNYJEhSZIkaagsMiRJkiQNlUWGJEmSpKGyyJAkSZI0VBYZkjRiSVYlOTPJuUm+0Dxder77+EiS7ZvXb5q27nvDaqskSYPwFraSNGJJbqiquzavPw2cXnfigVb9+5MkaRTsyZCkNct3gPsDJHlt07txbpIDm2UbJDk+yVnN8uc1y7+VZKck7wLWa3pGPt2su6H5M0ne27zvnL73PqF5/xeTXJDk083TdCVJWpClo26AJKknyVJgD+C/kjwKeBmwMxDg1CTfpvdU+iuq6i+a99ytfx9V9cYkB1TVDjNE7EPv6fKPADYDTktycrPukcBDgCuA7wKPBU4Z8keUJE0IezIkafTWS3ImsAL4BfBRYDfgy1V1Y1XdABwD/BlwDvDkJO9O8mdV9dt55OwGfKaqVlXVr4BvA49u1v2gqi6rqtuAM4Fth/LJJEkTyZ4MSRq930/veZjtaqWq+nGSHYE9gXck+UZVHT6ENtzc93oV/v8gSboT7MmQpDXTd4BnJlk/yQbAs4DvJNkSuKmqPgW8F9hxhvfemmTtWfb5vCRLkmwOPA74QUvtlyRNML+pkqQ1UFX9MMnHuL0I+EhVnZHkqcB7k9wG3Aq8eoa3LwfOTvLDqnpB3/IvA38KnAUU8Iaq+mWSB7X2QSRJE8lb2EqSJEkaKi+XkiRJkjRUFhmSJEmShsoiQ5IkSdJQWWRIkiRJGiqLDEmSJElDZZEhSZIkaagsMiRJkiQNlUWGJEmSpKH6/7e43AT4po2RAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 1080x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# compare saliency heatmaps for true positive examples and cases where model thought incorreclty that sequence was negative \n",
    " \n",
    "plt.rcParams[\"figure.figsize\"] = 15,5\n",
    "sns.heatmap([grads_over_letters_tp[letter]/counts_per_positon_tp[letter] for letter in new_alph], cmap='jet',yticklabels=[reverse_map[letter] for letter in new_alph])\n",
    "plt.xlabel('Position')\n",
    "plt.ylabel('Nucleotide Pairing')\n",
    "plt.title('Saliency Map: True Positives')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kcollins/anaconda/envs/deeptoehold/lib/python3.6/site-packages/ipykernel_launcher.py:4: RuntimeWarning: invalid value encountered in true_divide\n",
      "  after removing the cwd from sys.path.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1.0, 'Saliency Map: False Negatives')"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAxkAAAFNCAYAAABsY6I3AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvOIA7rQAAIABJREFUeJzt3Xvc7fWc///H0650Lh1HBzKSUwxJJU0OOYQUIcUw5bAxGsQ4zy/FfA1jNBkMNsph5KwmUxKKwqh2iQ7ClkMlnRVFtHv9/lifS6ur67r22lfr81l7Xetxv93Wba/PYX2e77Wuvdf+vK73+/35pKqQJEmSpGG5y6gbIEmSJGlhsciQJEmSNFQWGZIkSZKGyiJDkiRJ0lBZZEiSJEkaKosMSZIkSUNlkSFplZbkF0ke1zx/U5KPjLpNC0GSbZMs6GuYJ1mU5PdJ7jHqtkjSpLHIkNS6JLsl+W6S65Ncm+Q7SR6+ssepqrdX1YvaaON8JPlYkkqyz7T1/9GsP7CDNhyYZHlzMj31eF/bubO0ZdvmfR8/bf1nkvxzB/nf7v/Mq2p5Va1bVb9qO1uSdHsWGZJalWR94H+B9wIbAVsChwM3j7JdQ/QT4PlTC0lWA/YDftZhG/6vOZmeehzcYfZMdk2y04jbIEkaIYsMSW3bDqCqPt38ZvkPVXVyVf0QIMm9k5yS5JokVyf5VJINZzpQksOS/Hff8i5ND8lvk/wgyaP7tn0zyduaXpPfJTk5ySZ923fre+0lTY/Aw5NckWRR3377JvnBHO/vy8BuSe7WLO8J/BD4Td8x5nyPzZCwNya5MMl1SY5OsuZgH+/skuyd5NwkNyT5VZL/b459X9i043dJLk6yf9+2FyW5qGnbV5JsvYLodwH/bwXt+kHz2X87yfZ923Zs2vy7pgfk80kOa7ZtnOTEJFc1bflyki2bbe8EHgF8sOnNOTLJak3PyjZJHpnksiR36ct6VpJzmud3aYbj/az5GX1m6meaZO0kxzQ/v98mObP/75Ik6Y4sMiS17SfA8iQfT/KkvpPxKQH+FdgCuD+wNXDYig7anFyeAPwLvR6SfwK+mGTTvt2eAxwEbAas0exDknsCX6HXu7Ip8BDg3Ko6C7gGeELfMZ4HfGKOpvwR+B9g6qT8+TPsP8h7fC7wRODe9Aqzvwwvak5sd5ujDbP5fXPcDYGnAq9Mstf0nZrepiOAx1fVesAj6RVKJHkG8FpgH3qf1RnAMSvIfS/woP6iry/r4cCHgRcBGwNHAf+TZI0kdwWOAz5C72f6ReBpfS+/S/PaewD3BP4MvAegql4P/B/w0qY351XTor/b7P+ovnXP6XsvhwBPAXYHtqL32f1ns+0gYO1m/cbAP9D7uUuSZmGRIalVVXUDsBtQ9E4Qr0pyfJLNm+3LquprVXVzVV1F72T3UbMf8S/+Djixqk6sqlur6mvAUuDJffscXVU/qao/AJ+jV0xA7+Ty603vyp+r6pqqOrfZ9vHm2CTZiN6J/4pOqj8BPL/pnXgUvRPl/s9gkPf4vqq6pKqupdcLcEDf6zesqm/Pkb9LU4hMPXZpXndKVV3QfD4/AD4zQ+5fYoDtk6xZVZdX1YXN+pcCb6+qH1fVLfSKup2mehBmPFDVTcDbm32nWwz8V1Wd1fRsHdWsfzi94ubWqnpf83P5PHB233Gvqqpjm96wG5qMQf6uUFXVvP8DAJqf1RObdVPv801VdVlV/ZHekL5nNT0ffwY2AbZt2ry0qn4/SK4kTSqLDEmtq6ofVdWBVbUVsD293+gfCZBk82ZoymVJbgD+m94J3Yrck95J4F9OrukVM3fv2+c3fc9vAtZtnm/N7HMm/ht4apJ16M2tOL2qLl/B+/s2vd/yvxn436ao+YsB3+Mlfc9/Se8zGtT3mkJk6vG9JvcRzbCxq5JcT6/34A6fbXPCfgDwcuA3Sf43yXbN5nsC7+/7jK8GbqX3W/25fAi4R5InTVt/T+D1035ud6c3V2cL4NJp+//lc0mybpKPNEO/bgBOmen9zOEY4BlJVgeeAZxRVVN59wC+3Nem85r1mwEfA74OfK75Gb4jvbk3kqRZWGRI6lRVXUTvpG1qHP7b6f0W/UFVtT69XoQMcKhLgE9OO7lep6reMeBr7z1L+y6jN+xmX3pDpT45wPGgVzi8hpmHVg3yHvvnOdwD+PWAuXP5DL0hR1tX1Qb0hiHN+NlW1Veq6nH0TviX0SsSoPdZvXDa57xWVZ0xV3BV3Qy8lV5vRn/mJcDh0463dlV9DricXrHRr/9zeS1wL2Cn5nN87PTYFbRpaq7ME7n9UCnoFTePn9auNavqN1X1p6o6rKruT6+QfTq9YWiSpFlYZEhqVZL7JXlNkq2a5a3p/db8e80u69Eb/359MwTntQMeeqrH4Ynp3Q9hzSSPnspZgU8Bj0uyXzM5eOMkD+nb/gngdcCDgC8N2J7/BB4PnDbDtkHe48uTbNUM0Xoz8NkBc+eyHnBtVf2xGUK1/0w7Jbl7kqcmWRv4E3Ajvd4KgA8Cb05y/2bfDZM8c8D8jwHrA4/rW/dheu/14elZt8leB/g2sFqSlzU/l2cAD5v2fm4CrkuyMXDotLwrgL9eQZuOoTf/4hHAF/rWfxB4e5p7aiTZLMnezfPHJtm+GTp1A73hU7ciSZqVRYaktv0O2Bk4I8mN9IqL8+n91h96Y993AK6nN5F7oJP6qrqE3mTkNwFX0fsN+WsZ4HutuW/Ck5s2XAucC/xN3y7H0hvWc2wzv2CQ9lxbVd9oxv5PN8h7PAY4GbiY3lCuv8xnaK6W9LeDtGOalwH/muR39D6nz82y3yJ6n93l9Ca+70pv6BTNvIgjgM83Q5R+SK8nYIWaORxvoTeJe2rd95p2fQC4jt6FAf6u2XYzvV6Clzbb9gNO5LbLHR8BbNC08bv0Ju/3OxI4oBnydMQszTqGXg/I16rqur71RwAnAd9oPq/v0psnAr1hXF+iV2BcQG/o1Irm6UjSRMvM/x9K0mRL8jPgJVX19Q6yfgG8qIuscZPkbODIqhp02JokaRVgT4YkTdMM0yl6E4vVoWbI2+bNcKkXAvcDvjrqdkmSVo5FhiT1SfJNekN5Xl5Vjrvv3v3pDcn6LfAK4BlVdeVomyRJC1eSo5JcmeT8WbYnyX8mWZbkh0l2GOi4DpeSJEmSJlOS3eldnOQTVbX9DNufDPwjvbmMOwPvqaqdV3RcezIkSZKkCVVVp9G7CMps9qFXgFRz8Y4Nk9x9jv0BiwxJkiRJs9uS298w9lLueE+jO1hl71ia3ee+qVIbtjptWdeRbDGU+22tnEUs7zzzBJ7SeeZGJw505dHh2rD7yLW2v27FO7XgO+s/svPMT/D8zjP/49g3dp7Jmt1Hrv3IuX6J1Y6Pr//3nWfufdOXO8/8z7X/sfNMgK1vd07Qje+ya+eZ710y6K11huedLxnkfqHD9Xq+2HkmQNW+I8kdge5/qPNweLLS58eHwUuAxX2rllTVkqE1aharbJEhSZIk6c5pCoo7U1RcBmzdt7xVs25ODpeSJEmSxsBq83gMwfHA85urTO0CXF9Vlw/SVkmSJEmruNVbOGaSTwOPBjZJcinwlqmoqvogcCK9K0stA24CDhrkuBYZkiRJ0hho48S9qg5YwfYCXr6yx7XIkCRJksZAGz0ZbbHIkCRJksbAOJ24j1NbJUmSpIllT4YkSZKkoRqnE/dxaqskSZI0sezJkCRJkjRU43TiPk5tlSRJkibWOPVkdHbH7yRbJ3ltV3mSJEnSQrL6PB6j0mpPRpJNgWcBBwBbAMe2mSdJkiQtVOM0BGnobU2yHrAv8BxgO+BLwL2qaqsBXrsYWAzAth+Cuy8edvMkSZKksTROw6XaKIiuBM4E/hn4dlVVkqcP8sKqWgIsAcjuVAttkyRJksbSOPVktDEn443AXYH/At6Y5N4tZEiSJEkTZZzmZAy9yKiqI6tqF2CfZtVxwBZJXp9ku2HnSZIkSZNgtXk8RqW1q0tV1cVV9faqehCwI/AE4Ctt5UmSJElaNbR9damH0ruy1H7AL4Aj28yTJEmSFqqJnvjdDIk6oHlcDXwWSFU9ethZkiRJ0qQYp4nfbbT1IuB0YK+qWgaQ5JAWciRJkqSJMU49GW3MydgXuBw4NcmHk+wBpIUcSZIkaWKM08TvoWdX1XHAcUnWoXeFqVcBmyX5AHBsVZ087ExJkiRpoZv0ngwAqurGqjqmqp4KbAV8H3h9W3mSJEnSQjZO98nopBelqq6jdyfvJV3kSZIkSQvNpE/8liRJkjRkq4/RmXuqatRtmFFyePcN+8e3dB7JP93SfebVI/gb+uzuIzloBJknjSDzmSPIBHjmzZ1HHrjF0Z1nvp53dJ55vz1/2Xnm+096YeeZL//YRzvP/PmBd+88c+Pl13SeCbD+z//Ueebe236u88wvv/hZnWfWI7u/nk0OemfnmQBVrxtJ7giMxUWKblgrK31+vP4faiTvbYzqIUmSJGlyrb5o1C0YnEWGJEmSNAZWG6Mz9zFqqiRJkjS5xmlOxhg1VZIkSZpgDpeSJEmSNFRjdOY+Rk2VJEmSJtgYnbmPUVMlSZKkCTZGZ+53GXUDJEmSJC0sY1QPSZIkSRPMid+SJEmShmqMztzHqKmSJEnSBBujM/ehz8lIcvKwjylJkiRNvEXzeIxIG/XQpi0cU5IkSZpsY9ST0UZTN0iy72wbq+pLs21LshhY3FvaC9hx2G2TJEmSxtOkFxn0KoTMsK2AWYuMqloCLAFIDq8W2iZJkiSNpwm/utQvq+oFLRxXkiRJmlxj1JPRxs34ZurBIMluSd7fQp4kSZK08K02j8eItBH9vKknSR4KPAd4FvBz5hgqJUmSJGkOY9ST0UZT/5TkLcABwNXAZ4FU1WNayJIkSZImw4TPybgIOB3Yq6qWASQ5pIUcSZIkaXKMUU9GG3My9gUuB05N8uEkezDLPA1JkiRJAxqjORlDLzKq6riq2h+4H3Aq8CpgsyQfSPKEYedJkiRJE2GM7vjdRk8GAFV1Y1UdU1VPBbYCvg+8vq08SZIkaUGb5J6MmVTVdVW1pKr26CJPkiRJWnBaKjKS7Jnkx0mWJXnDDNvvkeTUJN9P8sMkT17RMTspMiRJkiStepIsAt4PPAl4AHBAkgdM2+2fgc9V1UOB/YH/WtFxLTIkSZKkcdDOnIydgGVVdXFV/Qn4DLDPtH0KWL95vgHw6xUddIwuhNWBb3YfufGRV3SeudY9/tB55uY/7f59nn38IzvPZGn3kY9+xUndhwL/e/NTOs/ch+M7z7zfEb/sPJMnACd3H6uF5ahtn9N55tM4juN4WqeZ9/7wBZ3mARw+kmtmbjaKUK1q5nHmnmQxsLhv1ZKqWtK3vCVwSd/ypcDO0w5zGHBykn8E1gEe10JTu1H1ls4z8+DOIyVpZidVp3Ev7zStceBHOo+8V+eJjPDqLs8dSerxPKvTvG3pvsiQRmYeZ+5NQbFkhTvO7QDgY1X17iSPAD6ZZPuqunW2F6yyRYYkSZKkPu2cuV8GbN23vFWzrt8LgT0Bqur/kqwJbAJcOdtBnZMhSZIkjYN25mScBdwnyb2SrEFvYvf08cm/AvYASHJ/YE3gqrkOak+GJEmSNA5aOHOvqluSHAx8lV5ZclRVXZDkrcDSqjoeeA3w4SSH0JsEfmBVzTmu1yJDkiRJGgctnblX1YnAidPWHdr3/EJgpa6oY5EhSZIkjYORXUhi5VlkSJIkSeNgjM7cx6ipkiRJ0gQbozP3MWqqJEmSNMHG6Mx9jJoqSZIkTTDnZEiSJEkaqjE6cx96U5NsCmzaXOqqf/0DgKuqas4bd0iSJEmawRgVGW3c8fu99G4zPt3GwHtayJMkSZIWvnbu+N2KNoqMbavqtOkrq+p04MFzvTDJ4iRLkyxdsmRJC02TJEmSxtRq83iMSBvR682xbfW5XlhVS4Cp6mLOW5VLkiRJWjW10ZOxLMmTp69M8iTg4hbyJEmSpIVvwnsyXgWckGQ/4Oxm3Y7AI4C9WsiTJEmSFr4xuoTt0HsyquqnwIOAbwHbAPcEvgm8CHjlsPMkSZKkiTDhPRlU1c3A0Ul2AA4A3gL8HPhiG3mSJEnSgjdGl7Bt4z4Z29ErLA4ArgY+C6SqHjPsLEmSJGliTHKRAVwEnA7sVVXLAJIc0kKOJEmSNDkmeU4GsC9wOXBqkg8n2QNICzmSJEnS5BijORltTPw+rqr2B+4HnErvalObJflAkicMO0+SJEmaCJNcZEypqhur6piqeiqwFfB94PVt5UmSJEkL2qJ5PEakk/qmqq6jdyfvJSvaV5IkSdIMJnzityRJkqRhG6Mz91W2qcm13Yd+Z6POI685d8vOMxfv8J7OM8/loZ1ncmT3kWzTfeQ3X7An677v6s5z1/2X5Z1ncmD3kSP5lnwIcNoIrpexe3Wf2bFlbN155lJ27DwT4EIe1nnmJWzNV/l4p5k/e9Pfd5o3OleOugFaFYzR1aVW2SJD0mBGUWBIkqQRGKMz9zFqqiRJkjTBxujMfYyaKkmSJE2wMTpzH6OmSpIkSRPMORmSJEmShmqMztxbuxmfJEmSpMk0RvWQJEmSNMHG6Mx9jJoqSZIkTTDnZEiSJEkaqjE6cx+jpkqSJEkTbIzO3FfY1CT/OcPq64GlVfU/w2+SJEmSpDsYoyJjkKtLrQk8BPhp83gwsBXwwiRHttg2SZIkSY1atPKPURmkHnow8MiqWg6Q5APA6cBuwHnTd06yG/DXVfWJZvkLwEbN5n+pqlOG0XBJkiRpkiyfR0/GqDo/BunJuBuwbt/yOsBGTdFx8wz7Hw4s7Vu+L/Ba4DDgdXMFJVmcZGmSpfDxAZomSZIkTYblq638Y1QGif434Nwk3wQC7A68Pck6wNdn2H/9qrqwb/mnVXU2QJJ/nSuoqpYAS3r7XlsDtE2SJEmaCLcsWvn7aN+1hXYMYoVFRlV9NMmJwE7NqjdV1a+b56+d4SUbTnv9vn2Lm8+rlZIkSdKEW77a+Mz8HrSldwGuavbfNsm2VXXaLPtelOQpVXVC/8okewE/nn9TJUmSpMm1fNH43I1vkEvYvhN4NnABcGuzuoDZioxDgBOSPBM4p1n3MGBXYK871VpJkiRpQi0fo1t+D9KT8TTgvlU10yTvO6iqZUkeDDwXeCC9guRbwFHAK4GXz7OtkiRJ0sS6ZYEVGRcDqzPzlaRm1BQkRyXZATgAeAvwc+CL82mkJEmSNOmWj9Hd+AZp6U30ri71DfoKjap6xUw7J9mOXmFxAHA18FkgVfWYO99cSZIkaTIttOFSxzePQV1E72Z9e1XVMoAkh8yjbZIkSZIaC6rIqKqVvSvevsD+wKlJTgI+Q+/+GpIkSZImwKx39EjyuebP85L8cPpjttdV1XFVtT9wP+BU4FXAZkk+kOQJw34DkiRJ0iRYzqKVfgwiyZ5JfpxkWZI3zLLPfkkuTHJBkmNWdMy5ejJe2fw5r8vOVtWNwDHAMUnuBjwLeD1w8nyOJ0mSJE2yNq4ulWQR8H7g8cClwFlJjq+qC/v2uQ/wRuCRVXVdks1WdNxZi4yqurwJ/didnbRdVdcBS5qHJEmSpJXU0tWldgKWVdXFAEk+A+wDXNi3z4uB9zfn9FTVlSs66JwtrarlSW5NskFVXT/vps/HtzbqNA7gCbuuzPz24VjELZ1nfpUndp65I2d3nnnm4zqPhO27j/z91zfh0Xuf1HnuS97+oc4zP8uzO8/kuO4j+Ub3kW84/DDg8E4zn8qXO80DWJv1Os/chl90ngnwuxG815FcXvNF1XnkPf+180hgrVGEahXT0sTvLYFL+pYvBXaets92AEm+AywCDquqOU8+Bvk2+D1wXpKvATdOrZztEraSujWKAkOSJHVvPkVGksXA4r5VS6pqZUcXrQbcB3g0sBVwWpIHVdVv53rBinypeUiSJEkakfnMyWgKirmKisuArfuWt2rW9bsUOKOq/gz8PMlP6BUdZ8120DYuYStJkiRpyFoakngWcJ8k96JXXOwPPGfaPsfRu9H20Uk2oTd86uK5DjprS5N8rqr2S3IecIcBj1X14JVrvyRJkqT5amNORlXdkuRg4Kv05lscVVUXJHkrsLSqjm+2PSHJhcBy4LVVdc1cx23tEraSJEmShqetO35X1YnAidPWHdr3vIBXN4+BzHkJ2+bPX650SyVJkiQNVVtFRhtmveP3lCS7JDkrye+T/CnJ8iQ3dNE4SZIkST23sGilH6MyyOyR99GbAPJ5YEfg+TTXypUkSZLUjZHci2aeVtiTAVBVy4BFVbW8qo4G9my3WZIkSZL6LWfRSj9GZZBy6KYkawDnJvk34HIGLE4kSZIkDceCmpMBPK/Z72B6d/zeGnjGbDsnGZ9+HEmSJGlMLJg5GUkeAmwLXFBVPwIOH+CYZwI7DKFtkiRJkhoLYk5GkkOBz9HrtTghyYsHPGaG0TBJkiRJ42mucujZwEOq6qYkGwMnAR8e4JibJpn1Rh1VdcRs25IsBhYD8JoPwd6LB4iTJEmSFr5xmpMxV5Fxc1XdBFBV1yQZdLL3ImBd5tGjUVVLgCUAOY1a2ddLkiRJC9VCKTL+OsnxzfMA9+5bpqr2nuV111bVW4fVQEmSJEmMdCL3ypqryNhn2vK/D3jMm+fZFkmSJEmzGKeJ37O2tKq+Nc9jjk+JJUmSJI2JhTJcar42nu/Eb0mSJEkzm/QiY94TvyVJkiTNbEEWGUnWnrra1Apc7sRvSZIkabjGaeL3Ci9Lm2TXJBcCFzXLf5Pkv+Z6ybAaJ0mSJKlnOaut9GNUBkn+D+CJwPEAVfWDJLvPsf8ew2iYJEmSpNssuOFSVXVJcrsOiuVz7HvtnW2UJEmSpNtbaEXGJUl2BSrJ6sArgR+12yxJkiRJ/cZpTsYgRcZLgfcAWwKXAScDL2+zUZIkSZJub0HcjG9KVV0NPLeDtkykzbmy88yvfGLfzjM3f37373OnN53GmW+fa/rQ8G2/91md5gFczcZswjWd536Il/ASPtRp5tZc0mkeABt0H8kGwK+7jXwen+w2EHjgVT/rPBPg+5vev9O8Lbr+YTZGMaxiFO/1t3+9YeeZv+w8EcDR6Fogw6WSvBeo2bZX1StaadHU8bs9N2zs3XniQXyw88xJUm/qNu9B3cYBjKTAmLI/x3aa913e1WneSB0+69dvS7btOG90HsqFneb9is07zZO0cI1TkTHXJWyXAmcDawI7AD9tHg8B1mi/aZIkSZKmLGfRSj9GZdaejKr6OECSlwG7VdUtzfIHgdO7aZ4kSZKkcTPI7JG7Aetz22DAdZt1kiRJkjqy0K4u9Q7g+0lOpXc3792Bw9pslCRJkqTbW2hXlzo6yVeAnZtVr6+q37TbLEmSJEn9xmni91xXl7pfVV2UZIdm1dS1I7dIskVVndN+8yRJkiTBAikygFcDi4F3z7CtgMe20iJJkiRJd7Ag5mRU1eLm6ZOq6o/925Ks2WqrJEmSJN3OOM3JmOs+GVO+O+A6SZIkSS1ZEPfJSPJXwJbAWkkeSu/KUtC7nO3aHbRNkiRJUmOhzMl4InAgsBVwRN/63wFvmu1FSbYFNq+q70xb/0jgN1X1s3m3VpIkSZpQC6LIaO74/fEkz6iqL67EMY8E3jjD+huabU9duSZKkiRJGqeJ34PMyfhGkiOSLG0e706ywRz7b15V501f2azbZq6gJIuncpYsWTJA0yRJkqTJsJzVVvoxKoMkfxQ4H9ivWX4ecDSw7yz7bzjHsdaaK6iqlgBT1UUN0DZJkiRpIiyI4VJ97l1Vz+hbPjzJuXPsvzTJi6vqw/0rk7wIOHs+jZQkSZIm3UIrMv6QZLeq+jb8ZQL3H+bY/3XAZ5M8l9uKih2BNYCn35nGSpIkSZNqnOZkDFJkvIzeBPCpeRjX0bvq1Gz+p6p2SPJY4IHNuhOq6pT5N1OSJEmabON0M74VtrSqzgX+Jsn6zfINK3hJmv1OASwsJEmSpCFYUMOlkmwOvB3YoqqelOQBwCOq6qOzvGTTJK+e7XhVdcRs2yRJkiSNv0EuYfsx4KvAFs3yT4BXzbH/ImBdYL1ZHpIkSZJW0nIWrfRjVAYZ2LVJVX0uyRsBquqWJMvn2P/yqnrrcJonSZIkCRbexO8bk2xMc9+KJLsA18+xf4bRMEmSJEm3GaeJ34MMl3o1cDxw7yTfAT4B/OMc++8xjIZJkiRJuk1bw6WS7Jnkx0mWJXnDHPs9I0kl2XFFxxzk6lLnJHkUcF96vRQ/rqo/z7H/tSs6piRJkqSV08YciySLgPcDjwcuBc5KcnxVXThtv/WAVwJnDHLcWYuMJPvOsmm7JFTVlwZquSRJkqQ7raWJ3DsBy6rqYoAknwH2AS6ctt/bgHcCrx3koHP1ZDx1jm0FWGRIkiRJHWlp4veWwCV9y5cCO/fvkGQHYOuqOiHJnSsyquqg+bRyWPLwEYTeOILMo1/afea53Ueeee7u7HTEad0Hd+z8h4/gL+7W3UcCPPpLJ/EYTuo0czvW6DQP4CUvOZIPfWiuq3a35EPdXkNj6Uv26zQPYOmmD+Pvr/pcp5mf2fRpwNM7zYRd2ZXvdpw5Gns9/uvdhz6o+0gOeSmH/UfX17nZqOM8rYrmM/E7yWJgcd+qJVW1ZCVefxfgCODAlckd5GZ8bwf+rap+2yzfDXhNVf3zygRJZ7D7qJsgzc9Lqtu8jguMkdq048+28wKj5x5cMYJMtekt1e3f3cPy3k7ztGqaz3CppqCYq6i4jNv/ynKrZt2U9YDtgW8mAfgr4Pgke1fV0tkOOsjVpZ40VWA0Db0OePIAr5MkSZI0JC1dXeos4D5J7pVkDWB/eleWBaCqrq+qTapqm6raBvgeMGeBAYPdJ2NRkrtW1c0ASdYC7jpIiyVJkiQNRxtzMpobbR8MfBVYBBxVVRckeSuwtKqOn/sIMxukyPgU8I0kRzfLBwEfn0+YJEmSpPlp62Z8VXUicOK0dYfOsu+jBznmIPfJeGeSH3LbTfbeVlVfHeTgkiRJkoajpUvYtmKgcqiqvgJ8peW2SJIkSZrFgioykvyO3n0xANYAVgdurKr122yYJEmSpNssv3UeRcYgl3lqwSDDpdabep524tg7AAAWCElEQVTedav2AXZps1GSJEmSbu+WW+ZRZHR/iylgJWub6jkOeGJL7ZEkSZI0g+W3rLbSj1EZZLjUvn2LdwF2BP7YWoskSZIkjbVBypun9j2/BfgFvSFTkiRJkjqyfD7DpUZkkDkZB63MAZPsA2xVVe9vls8ANm02v66qvrDSrZQkSZIm3IIoMpLMeAOORlXV22bZ9jp6tyOfclfg4cA6wNGARYYkSZK0km758wIoMoAbZ1i3DvBCYGNgtiJjjaq6pG/521V1DXBNknXmakySxcBiAO75Idh08Vy7S5IkSRPj1uWjm8i9smZtaVW9e+p5kvWAVwIHAZ8B3j3b64C7TTvOwX2LmzKHqloCLAHIw/9ybw5JkiRJYzRcas5L2CbZKMm/AD+kV5DsUFWvr6or53jZGUlePMOxXgKceadaK0mSJE2qWxat/GNE5pqT8S5gX3o9Cw+qqt8PeMxDgOOSPAc4p1n3MHpzM552J9oqSZIkTa5bMuoWDGyugV2vAW4G/hl4c+9m3wCE3sTv9Wd53UlVtUOSPYAHNOtOqKpThtFgSZIkaSLdMuoGDG6uORkrdTfwPmle/w3gG/M8hiRJkqR+C6HIuBM2TfLq2TZW1REtZEqSJEkL24QXGYuAdWl6NCRJkiQNwZ9H3YDBtVFkXF5Vb23huJIkSdLkWj7qBgyujSLDHgxJkiRp2CZ8uNQeLRxTkiRJmmyTXGRU1bXDPqYkSZI08Sa5yJAkSZLUgjEqMlJVo27DjPIAOm/YThee1nUkj+HUzjOP5qDOM/+Jf+8883X/8J+dZ7Jb95E8+uYRhMK3tti988wP8ZLOMz/1ry/sPJN1uo889BVv6jzzII7uPPNeZ1zeeeahO3f/2QL8jvU6z1w0gjOg43h655nPy/adZx7GYZ1nAlS9ZSS5IzAWc4pz4sqfH9eTR/Pe7MmQJEmSxsEY9WRYZEiSJEnjwCJDkiRJ0lBN+M34JEmSJA3bhN+MT5IkSdKwOVxKkiRJ0lBZZEiSJEkaKosMSZIkSUNlkSFJkiRpqCwyJEmSJA2VRYYkSZKkoRqj+2TcpY2DJtk0yQNmWP+AJJu2kSlJkiQtaMvn8RiRVooM4L3AJjOs3xh4T0uZkiRJ0sJ1yzweI9JWkbFtVZ02fWVVnQ48eLYXJVmcZGmSpVy3pKWmSZIkSWNojIqMtuZkrDfHttVn21BVS4AlAHkANexGSZIkSWNrjCZ+t9WTsSzJk6evTPIk4OKWMiVJkqSF68/zeIxIWz0ZrwJOSLIfcHazbkfgEcBeLWVKkiRJWgW0VWQU8ALgPsD2zbpvAR9mpPPcJUmSpDE1RmfRbRUZRwJvrKqj+1cmeVCz7akt5UqSJEkL0xjNyWiryNi8qs6bvrKqzkuyTUuZkiRJ0sJlkcGGc2xbq6VMSZIkaeGa9Dt+A0uTvHj6yiQv4raJ4JIkSZIGNUZ3/G7z6lLHJnkut7+61BrA01vKlCRJkhauSR8uVVVXALsmeQy3XV3qhKo6pY08SZIkacFrqchIsifwHmAR8JGqese07a8GXtS04CrgBVX1y7mO2VZPBgBVdSpwapsZkiRJ0kRoYU5GkkXA+4HHA5cCZyU5vqou7Nvt+8COVXVTkpcB/wY8e67jtjUnQ5IkSdIwtTMnYydgWVVdXFV/Aj4D7NO/Q1WdWlU3NYvfA7Za0UFb7cm4U350Q+eRZ568e+eZP3vcvTvPvGbZlp1nXrjdAzrPPPC/PsjH/uGl3Yau8J9cC35x1xGEwr9v8U+dZz5sFNeNeHf3kQAc2m3cW7/w9m4DgY8/c7/OM9n59M4jD+JojuagznNH4RK27jzzZ49/YOeZhz2uOOzr6TxXms9wqSSLgcV9q5ZU1ZK+5S2BS/qWLwV2nuOQLwS+sqLcVbfIkIag/qvbvJzWbZ4WsFdUt3lf8ISpTW/l/426CZ14Nh8fdRM685bq9t/oYTm80zytouZRZDQFxZIV7jiAJH9H72JOj1rRvhYZkiRJ0jho5z4Zl8HtuiG3atbdTpLHAW8GHlVVN6/ooBYZkiRJ0jho574XZwH3SXIvesXF/sBz+ndI8lDgQ8CeVXXlIAe1yJAkSZLGQQuXsK2qW5IcDHyV3iVsj6qqC5K8FVhaVccD7wLWBT6fBOBXVbX3XMe1yJAkSZLGQUv3yaiqE4ETp607tO/541b2mBYZkiRJ0jhoZ05GK7xPhiRJkqShsidDkiRJGgftTPxuhUWGJEmSNA5ampPRBosMSZIkaRxYZEiSJEkaqjGa+G2RIUmSJI2DSZ+TkWQ34K+r6hPN8heAjZrN/1JVp7SRK0mSJC1YYzRcqq1L2B4OLO1bvi/wWuAw4HWzvSjJ4iRLkyyFo1tqmiRJkjSGbpnHY0TaGi61flVd2Lf806o6GyDJv872oqpaAizp7XdDtdQ2SZIkafw4J4MN+xeqat++xc1bypQkSZIWrjGak9HWcKmLkjxl+sokewE/bilTkiRJWrgcLsUhwAlJngmc06x7GLArsFdLmZIkSdLCNUYTv9u8hO0Lge2ABzbLpwEfYaw6eiRJkqRVhHMyOBJ4Y1Ud1b8yyYOabU9tKVeSJElamMboV/VtFRmbV9V501dW1XlJtmkpU5IkSVq4xujaq21N/N5wjm1rtZQpSZIkaRXQVpGxNMmLp69M8iLg7JYyJUmSJK0C2hou9Srg2CTP5baiYkdgDeDpLWVKkiRJWgW0UmRU1RXArkkeA2zfrD6hqk5pI0+SJEnSqqPNS9hSVacCp7aZIUmSJE2G+VzDdvWht2IQqVo1p6lni+7nz2/162VdR/K7P63beeb1v/irzjP32O6EzjO/8dk73HS+fdt0Hwlwl21u7DzzHZu/sfPMM9i588wvXPR3nWeO4mZLL9/+3Z1nPodjOs/cnCs6z7yQB3SeCbCMbTvPfM3F7+s8kzX/1Hlk/dOanWfm00d3nllbHNR5Zs76Y+eZALXFXTOS4JWU/GGlz4+r1hrJe2u1J0NS+0ZRYEiSpFGYT0/GaC7sapEhSZIkjYURdHXPk0WGJEmSNBbm05MxGhYZkiRJ0liwyJAkSZI0VA6XkiRJkjRU9mRIkiRJGip7MiRJkiQNlT0ZkiRJkobKngxJkiRJQ2VPhiRJkqShsidDkiRJ0lBNeE9GkgcC966q45vl/wA2aDa/r6rOaSNXkiRJWrjGpyfjLi0d9x3A1X3LTwROAE4FDp3tRUkWJ1maZCk3LWmpaZIkSdI4+vM8HqPR1nCpu1fVd/uWb6iqLwIkeclsL6qqJcASgGxBtdQ2SZIkSS1qq8hYr3+hqnbpW9yspUxJkiRpARuf4VJtFRm/TrJzVZ3RvzLJLsCvW8qUJEmSFrAJn/gNvB74bJKPAVOTvB8GHAjs11KmJEmStIBNeJFRVWcm2Rk4mF5hUcAFwE5VdWUbmZIkSdLCNuHDpZLsA2xVVYc2y2fS68l4XpLXVdUX2siVJEmSFq4J78kAXgfs37e8Br0iY13gaMAiQ5IkSVopE96TAaxRVZf0LX+7qq4Frk2yTkuZkiRJ0gJmT8bd+heq6uC+xU1bypQkSZIWsPHpyWjrjt9nJHnx9JXNjfjObClTkiRJWsDaueN3kj2T/DjJsiRvmGH7XZN8ttl+RpJtVnTMtnoyDgGOS/Icbn8J27sCT2spU5IkSVrAht+TkWQR8H7g8cClwFlJjq+qC/t2eyFwXVVtm2R/4J3As+c6bluXsL0S2DXJY4EHNqtPqKpT2siTJEmSFr5W5mTsBCyrqosBknwG2AfoLzL2AQ5rnn8BeF+SVFXNdtC2ejIAaIoKCwtJkiTpTmtlTsaWQP8Fmy4Fdp5tn6q6Jcn1wMbA1bMetaoW3ANYbKaZZq66uWaaaeaqmzlJ79VMMyfhASwGlvY9Fk/b/kzgI33LzwPeN22f8+ndA29q+WfAJnPltjXxe9QWm2mmmat0rplmmrnqZo4q10wzxzFzlVdVS6pqx77Hkmm7XAZs3be8VbNuxn2SrAZsAFwzV+5CLTIkSZIkrdhZwH2S3CvJGvRuqH38tH2OB/6+ef5M4JRqujRm0+qcDEmSJEmrrurNsTgY+CqwCDiqqi5I8lZgaVUdD3wU+GSSZcC19AqROS3UImN6N5CZZpq5auWaaaaZq27mqHLNNHMcMxeEqjoROHHaukP7nv8ReNbKHDMr6OmQJEmSpJXinAxJkiRJQ7WgiowV3RK9pcyjklyZ5Pwu8prMrZOcmuTCJBckeWUHmWsmOTPJD5rMw9vO7MtelOT7Sf63o7xfJDkvyblJlnaUuWGSLyS5KMmPkjyi5bz7Nu9v6nFDkle1mdnkHtL8/Tk/yaeTrNlB5iubvAvafI8zfRck2SjJ15L8tPnzbh1kPqt5r7cm2XGYeXNkvqv5u/vDJMcm2bCDzLc1eecmOTnJFm1n9m17TZJKsknbmUkOS3JZ37/VJ7ed2az/x+ZnekGSf2s7M8ln+97jL5KcO8zMOXIfkuR7U9/3SXbqIPNvkvxf8//Ml5OsP8S8Gc8POvgumi23te+jOTJb/T7SShj1tXuHeA3gRfSu2fvXwBrAD4AHdJC7O7ADcH6H7/XuwA7N8/WAn7T9XoEA6zbPVwfOAHbp6P2+GjgG+N+O8n7BCq793ELmx4EXNc/XADbsMHsR8Bvgni3nbAn8HFirWf4ccGDLmdvTu7b32vTmoH0d2LalrDt8FwD/Bryhef4G4J0dZN4fuC/wTWDHjt7nE4DVmufv7Oh9rt/3/BXAB9vObNZvTW9y5C+H/T0xy/s8DPinYf8cV5D5mObfyl2b5c26+Gz7tr8bOLSj93oy8KTm+ZOBb3aQeRbwqOb5C4C3DTFvxvODDr6LZstt7ftojsxWv498DP5YSD0Zf7klelX9CZi6JXqrquo0erPsO1NVl1fVOc3z3wE/oncC12ZmVdXvm8XVm0frE3qSbAU8BfhI21mjkmQDev8RfRSgqv5UVb/tsAl7AD+rql92kLUasFZ619heG/h1y3n3B86oqpuq6hbgW8C+bQTN8l2wD70CkubPp7WdWVU/qqofDzNngMyTm88X4Hv0rrHeduYNfYvrMOTvozm+2/8DeN2w81aQ2ZpZMl8GvKOqbm72ubKDTACSBNgP+PQwM+fILWCqJ2EDhvydNEvmdsBpzfOvAc8YYt5s5wdtfxfNmNvm99Ecma1+H2lwC6nImOmW6K2eeK8KkmwDPJRez0LbWYuaLuwrga9VVeuZwJH0/kO/tYOsKQWcnOTsJF3c2OdewFXA0ekNC/tIknU6yJ2yPy38hz5dVV0G/DvwK+By4PqqOrnl2POBv02ycZK16f2mcusVvGaYNq+qy5vnvwE27zB7VF4AfKWLoCT/L8klwHOBQ1e0/xDy9gEuq6oftJ01zcHN0I+jhj3MZRbb0ft3c0aSbyV5eAeZU/4WuKKqftpR3quAdzV/j/4deGMHmRdw2y9Bn0VL30nTzg86+y7q8rxkgMzOvo90RwupyJg4SdYFvgi8atpv9VpRVcur6iH0fiuwU5Lt28xLshdwZVWd3WbODHarqh2AJwEvT7J7y3mr0etO/0BVPRS4kV53duvSu+nO3sDnO8i6G73/WO8FbAGsk+Tv2sysqh/R6y4/GTgJOBdY3mbmHG0pOuj9G6UkbwZuAT7VRV5Vvbmqtm7yDm4zqylS30QHxcw0HwDuDTyEXnH+7g4yVwM2AnYBXgt8rulh6MIBdPBLjz4vAw5p/h4dQtOj3LIXAP+Q5Gx6w3z+NOyAuc4P2vwu6vq8ZK7Mrr+PdEcLqcgY5JboC0aS1en9o/pUVX2py+xmKM+pwJ4tRz0S2DvJL+gNf3tskv9uOXPqN+5TQwSOpTcUr02XApf29Qx9gV7R0YUnAedU1RUdZD0O+HlVXVVVfwa+BOzadmhVfbSqHlZVuwPX0Ru325UrktwdoPlzqMNOViVJDgT2Ap7bnMR06VMMccjJLO5Nr0D+QfOdtBVwTpK/ajO0qq5ofsFzK/Bh2v8+gt530peaYbJn0utJHuok95k0wyj3BT7bdlafv6f3XQS9X7a0/vlW1UVV9YSqehi9gupnwzz+LOcHrX8XjeK8ZLbMEX8fqbGQioxBbom+IDS/Ufoo8KOqOqKjzE2nrtCQZC3g8cBFbWZW1Ruraquq2obez/OUqmr1N99J1kmy3tRzehPIWr1yWFX9BrgkyX2bVXsAF7aZ2afL3xr+CtglydrN3+E96I2hbVWSzZo/70HvBOaYtjP7HE/vJIbmz//pMLszSfakN6xx76q6qaPM+/Qt7kP730fnVdVmVbVN8510Kb1Jp79pM3fqxLDxdFr+PmocR2/yN0m2o3cxiqs7yH0ccFFVXdpB1pRfA49qnj8WaH2YVt930l2AfwY+OMRjz3Z+0Op30YjOS2bMHMX3kWaxMrPEV/UHvfHWP6H3W4E3d5T5aXpd2H+m95/OCzvI3I1eV+cP6Q3/OBd4csuZDwa+32SeTwtX/lhB/qPp4OpS9K5O9oPmcUGHf48eAixtPt/jgLt1kLkOcA2wQYc/x8PpnQyeD3yS5uo1LWeeTq9o+wGwR4s5d/guADYGvkHvxOXrwEYdZD69eX4zcAXw1Q4yl9GbEzf1fTTsKz3NlPnF5u/RD4Ev05vw2WrmtO2/YPhXl5rpfX4SOK95n8cDd+8gcw3gv5vP9xzgsV18tsDHgJcOM2uA97obcHbz/XAG8LAOMl9J71zlJ8A7aG6MPKS8Gc8POvgumi23te+jOTJb/T7yMfjDO35LkiRJGqqFNFxKkiRJ0irAIkOSJEnSUFlkSJIkSRoqiwxJkiRJQ2WRIUmSJGmoLDIkacSSLE9ybpLzk3y+ubv0yh7jI0ke0Dx/07Rt3x1WWyVJGoSXsJWkEUvy+6pat3n+KeDsuhM3tOo/niRJo2BPhiStWk4HtgVI8uqmd+P8JK9q1q2T5IQkP2jWP7tZ/80kOyZ5B7BW0zPyqWbb75s/k+RdzevO63vto5vXfyHJRUk+1dxNV5KkeVlt1A2QJPUkWQ14EnBSkocBBwE7AwHOSPItenel/3VVPaV5zQb9x6iqNyQ5uKoeMkPEvvTuLv83wCbAWUlOa7Y9FHgg8GvgO8AjgW8P+S1KkiaEPRmSNHprJTkXWAr8CvgosBtwbFXdWFW/B74E/C1wHvD4JO9M8rdVdf1K5OwGfLqqllfVFcC3gIc3286sqkur6lbgXGCbobwzSdJEsidDkkbvD9N7HmYbrVRVP0myA/Bk4F+SfKOq3jqENtzc93w5/v8gSboT7MmQpFXT6cDTkqydZB3g6cDpSbYAbqqq/wbeBewww2v/nGT1WY757CSLkmwK7A6c2VL7JUkTzN9USdIqqKrOSfIxbisCPlJV30/yROBdSW4F/gy8bIaXLwF+mOScqnpu3/pjgUcAPwAKeF1V/SbJ/Vp7I5KkieQlbCVJkiQNlcOlJEmSJA2VRYYkSZKkobLIkCRJkjRUFhmSJEmShsoiQ5IkSdJQWWRIkiRJGiqLDEmSJElDZZEhSZIkaaj+fwe151Q7e0/dAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 1080x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# compare saliency heatmaps for true positive examples and cases where model thought incorreclty that sequence was negative \n",
    " \n",
    "plt.rcParams[\"figure.figsize\"] = 15,5\n",
    "sns.heatmap([grads_over_letters_fn[letter]/counts_per_positon_fn[letter] for letter in new_alph], cmap='jet',yticklabels=[reverse_map[letter] for letter in new_alph])\n",
    "plt.xlabel('Position')\n",
    "plt.ylabel('Nucleotide Pairing')\n",
    "plt.title('Saliency Map: False Negatives')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Recall: 0.5235294117647059\n",
      "Precision: 0.717741935483871\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import recall_score, precision_score\n",
    "\n",
    "print('Recall:', recall_score(y_true_labels, y_pred_labels))\n",
    "print('Precision:', precision_score(y_true_labels, y_pred_labels))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Guide seq:  GACCCCCTCCACCCCGCCTCCGG\n",
      "Off-target seq: GACCCCCTTCACCCCACCTATGG\n",
      "Encoded seq:  EKPPPPPLMPKPPPPYPPLZMEE\n",
      "New alphabet mapping:  {'AA': 'K', 'AT': 'R', 'AC': 'Z', 'AG': 'Y', 'TT': 'L', 'TC': 'M', 'TG': 'J', 'CC': 'P', 'CG': 'H', 'GG': 'E'}\n"
     ]
    }
   ],
   "source": [
    "sample_idx = 70\n",
    "print('Guide seq: ', test_df['On-Target Site'][sample_idx])\n",
    "print('Off-target seq:', test_df['Off-Target Site'][sample_idx])\n",
    "print('Encoded seq: ', test_df['Encoded Seqs'][sample_idx])\n",
    "print('New alphabet mapping: ', new_alph_map)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         (None, 23)                0         \n",
      "_________________________________________________________________\n",
      "embedding_1 (Embedding)      (None, 23, 5)             50        \n",
      "_________________________________________________________________\n",
      "flatten_1 (Flatten)          (None, 115)               0         \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 115)               0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 50)                5800      \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 50)                0         \n",
      "_________________________________________________________________\n",
      "preds (Dense)                (None, 2)                 102       \n",
      "=================================================================\n",
      "Total params: 5,952\n",
      "Trainable params: 5,952\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.6.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}