MNIST_dumb_Autoencoder/day1.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import polars as pl\n",
    "\n",
    "splits = {'train': 'mnist/train-00000-of-00001.parquet', 'test': 'mnist/test-00000-of-00001.parquet'}\n",
    "df = pl.read_parquet('hf://datasets/ylecun/mnist/' + splits['train'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "b'\\x89PNG\\r\\n\\x1a\\n\\x00\\x00\\x00\\rIHDR\\x00\\x00\\x00\\x1c\\x00\\x00\\x00\\x1c\\x08\\x00\\x00\\x00\\x00Wf\\x80H\\x00\\x00\\x00\\xf6IDATx\\x9cc`\\x18\\xdc@e\\xd2\\xbf\\xff\\xbf\\xe3\\xd9\\xb0I1\\'\\xbe\\xfa\\xf7\\xef\\xc5\\xbf\\x7f\\xcaX\\xe4\"{\\xfe\\xfd[\\xe7\\xacx\\xe2\\xdf>L\\xb9\\x9c\\xbf\\xff^\\x1b320\\xf0\\xfc\\xfbk\\x81.\\xc7s\\xe3\\xdf\\x17s\\x06\\x06\\x06\\x06\\xb6\\xab\\xff,!BLpI^5\\x86\\t\\'\\x19\\x18\\x18\\x18~]\\x86\\t!$E\\x18\\xbe\\xceC3\\x0c!\\x19\\xcc\\xb0\\xea\\x1e.I\\xe1\\x14\\x863\\x10\\x16\\xbb\\r\\x86\\xa4\\xba\\x14\\xc3;\\x08\\x8bY\\xea\\xc774#\\xac\\xfe\\xfd\\xe3\\x84\\xb0\\xb8\\xfe]\\xc2\\xb0\\x13\\x0e\\xe2\\x19:\\xd1\\x85XoCuJ<\\xfb\\'\\x81\\xae\\xf3\\xf7_(\\xc3F\\xe2\\xef\\x7f\\x0c\\xd3n\\xfcSa```\\x10\\xbb\\xf0;\\x0e\\xd3\\xaa\\xc6\\x7f\\x15\\x0c\\x0c\\x0c\\xcc=\\xff\\x9eb\\xca1\\x04\\xff\\xbb\\'\\xc8\\xc0\\x10\\xfd\\xef\\x8d)\\x16I\\xa1\\xd7\\xff\\xdaD\\xcb\\xdf\\xfe\\xab\\xc1\"\\xc7\\xc0`\\xfa\\xf3\\xdf\\xab?\\xff\\xd60c\\x95d(y\\xfb\\xef_\\xab\\x08v9j\\x02\\x00\\x03\\xd3Q2S\\xd8\\\\5\\x00\\x00\\x00\\x00IEND\\xaeB`\\x82'\n"
     ]
    },
    {
     "data": {
      "image/jpeg": "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",
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAABwAAAAcCAAAAABXZoBIAAAA+klEQVR4AWNgGNxAZdK//7/j2bA5kjnx1b9/L/79U8YiGdnz7986Z8UT//ZhSub8/ffamJGBgeffXwt0WZ4b/76YgwTZrv6zhEgywdXwqjFMOAni/boME0NIijB8nQcThdIIyWCGVfdwSQqnMJyByLHbwNTAdapLMbyDCDJL/fgGk4bSVv/+cUKYXP8uQcXgOhFq4xk6ERwIi/U2VKfEs38S6Dp//4WK2Ej8/Y+uk+HGPxWQmNiF33EYcgyN/yqAgsw9/55iyjEE/7snyMAQ/e+NKRZJodf/2kTL3/6rwSLHwGD689+rP//WMGOVZCh5++9fqwh2OWqKAgAD01EykXgccgAAAABJRU5ErkJggg==",
      "text/plain": [
       "<PIL.PngImagePlugin.PngImageFile image mode=L size=28x28>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from PIL import Image\n",
    "import IPython.display as display\n",
    "import io\n",
    "\n",
    "print(df[\"image\"][1000][\"bytes\"])\n",
    "image_data = df[\"image\"][1000][\"bytes\"]\n",
    "\n",
    "image = Image.open(io.BytesIO(image_data))\n",
    "image.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'polars.series.series.Series'>\n",
      "shape: (60_000, 3)\n",
      "┌─────────────────────────────────┬───────┬─────────────────────────────────┐\n",
      "│ image                           ┆ label ┆ tensor                          │\n",
      "│ ---                             ┆ ---   ┆ ---                             │\n",
      "│ struct[2]                       ┆ i64   ┆ object                          │\n",
      "╞═════════════════════════════════╪═══════╪═════════════════════════════════╡\n",
      "│ {b\"\\x89PNG\\x0d\\x0a\\x1a\\x0a\\x00… ┆ 5     ┆ tensor([0.0000, 0.0000, 0.0000… │\n",
      "│ {b\"\\x89PNG\\x0d\\x0a\\x1a\\x0a\\x00… ┆ 0     ┆ tensor([0.0000, 0.0000, 0.0000… │\n",
      "│ {b\"\\x89PNG\\x0d\\x0a\\x1a\\x0a\\x00… ┆ 4     ┆ tensor([0.0000, 0.0000, 0.0000… │\n",
      "│ {b\"\\x89PNG\\x0d\\x0a\\x1a\\x0a\\x00… ┆ 1     ┆ tensor([0.0000, 0.0000, 0.0000… │\n",
      "│ {b\"\\x89PNG\\x0d\\x0a\\x1a\\x0a\\x00… ┆ 9     ┆ tensor([0.0000, 0.0000, 0.0000… │\n",
      "│ …                               ┆ …     ┆ …                               │\n",
      "│ {b\"\\x89PNG\\x0d\\x0a\\x1a\\x0a\\x00… ┆ 8     ┆ tensor([0.0000, 0.0000, 0.0000… │\n",
      "│ {b\"\\x89PNG\\x0d\\x0a\\x1a\\x0a\\x00… ┆ 3     ┆ tensor([0.0000, 0.0000, 0.0000… │\n",
      "│ {b\"\\x89PNG\\x0d\\x0a\\x1a\\x0a\\x00… ┆ 5     ┆ tensor([0.0000, 0.0000, 0.0000… │\n",
      "│ {b\"\\x89PNG\\x0d\\x0a\\x1a\\x0a\\x00… ┆ 6     ┆ tensor([0.0000, 0.0000, 0.0000… │\n",
      "│ {b\"\\x89PNG\\x0d\\x0a\\x1a\\x0a\\x00… ┆ 8     ┆ tensor([0.0000, 0.0000, 0.0000… │\n",
      "└─────────────────────────────────┴───────┴─────────────────────────────────┘\n"
     ]
    }
   ],
   "source": [
    "# Convert images to tensors that can be trained on\n",
    "import numpy as np\n",
    "from torch import tensor\n",
    "import torch\n",
    "\n",
    "dataset = []\n",
    "\n",
    "def transform_image(image):\n",
    "    image = Image.open(io.BytesIO(image[\"bytes\"]))\n",
    "    data = np.array(image)\n",
    "    return tensor(data, dtype=torch.float32).flatten()/256\n",
    "\n",
    "print(type(df[\"image\"]))\n",
    "\n",
    "df = df.with_columns([\n",
    "    pl.Series(\"tensor\", [transform_image(img) for img in df['image'].to_list()])\n",
    "])\n",
    "\n",
    "# df[\"tensor\"] = df[\"image\"].apply(lambda x: transform_image(x[\"bytes\"]))\n",
    "print(df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "784"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df[\"tensor\"][0].size()[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "MNIST database of handwritten digits. 60000 images for training and 10000 for evaluation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch import nn\n",
    "from torch import functional as F #For training our model\n",
    "from torch import optim\n",
    "from IPython.display import clear_output, display\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "class my_network(nn.Module):\n",
    "    def __init__(self, feature_size, coding_layer):\n",
    "        super().__init__()\n",
    "        self.activation_function = nn.ReLU()\n",
    "        self.encoding1 = nn.Linear(in_features=feature_size, out_features=coding_layer*5, dtype=torch.float32)\n",
    "        self.encoding2 = nn.Linear(in_features=coding_layer*5, out_features=coding_layer, dtype=torch.float32)\n",
    "        self.decoding1 = nn.Linear(in_features=coding_layer, out_features=coding_layer*5, dtype=torch.float32)\n",
    "        self.decoding2 = nn.Linear(in_features=coding_layer*5, out_features=feature_size, dtype=torch.float32)\n",
    "\n",
    "\n",
    "    def forward(self, features):\n",
    "        x = self.encoding1(features)\n",
    "        # x = self.activation_function(x)\n",
    "        x = self.encoding2(x)\n",
    "        # x = self.activation_function(x)\n",
    "        x = self.decoding1(x)\n",
    "        # x = self.activation_function(x)\n",
    "        x = self.decoding2(x)\n",
    "        # x = self.activation_function(x)\n",
    "        return x\n",
    "\n",
    "def train(epochs: int):\n",
    "    network = my_network(df[\"tensor\"][0].size()[0], 10)\n",
    "\n",
    "    loss_function = nn.MSELoss() # Since we're working with softmax\n",
    "    optimizer = optim.Adam(network.parameters(), lr=0.001)\n",
    "\n",
    "    for epoch in range(0, epochs):\n",
    "        epoch_loss = 0\n",
    "        for image in df[\"tensor\"].sample(150):\n",
    "            clear_output(wait=True)\n",
    "\n",
    "            # print(image)\n",
    "            optimizer.zero_grad()\n",
    "            output = network(image)\n",
    "\n",
    "            # print(output)\n",
    "            # print(image)\n",
    "            loss = loss_function(output, image)\n",
    "            epoch_loss = epoch_loss+ loss\n",
    "            loss.backward()\n",
    "            print(loss)\n",
    "            optimizer.step()\n",
    "            fig, ax = plt.subplots(1, 1, figsize=(5, 5))\n",
    "\n",
    "            ax.imshow(output.detach().numpy().reshape((28, 28))*256, cmap='gray')\n",
    "                        # Render the updated figure\n",
    "            display(fig)\n",
    "            \n",
    "            # Close the figure to avoid redundancy\n",
    "            plt.close(fig)\n",
    "            # break\n",
    "        print(f'Epoch {epoch+1}/{epochs}, Loss: {epoch_loss/len(df[\"tensor\"])}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor(0.0382, grad_fn=<MseLossBackward0>)\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 500x500 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 10/10, Loss: 9.386472083861008e-05\n"
     ]
    }
   ],
   "source": [
    "train(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'model_graph.png'"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from torchview import draw_graph\n",
    "\n",
    "model_graph = draw_graph(my_network(df[\"tensor\"][0].size()[0], 10), input_size=df[\"tensor\"][0].size(), expand_nested=True)\n",
    "model_graph.visual_graph.render(\"./model_graph\", cleanup=True, format='png')"
   ]
  },
  {
   "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": "3.10.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}