Generative_MNIST_using_AutoEncoders(Sparse).ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "authorship_tag": "ABX9TyMYhoCSedvrTLhgdrTGiZFl",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/Ashu-00/AutoEncodersAndGenerators/blob/main/Generative_MNIST_using_AutoEncoders(Sparse).ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "1-jIrHUwxq3Z"
      },
      "outputs": [],
      "source": [
        "import torch\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F\n",
        "import torch.optim as optim\n",
        "import torchvision"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "dataset=torchvision.datasets.MNIST('data',train=True,download=True,transform=torchvision.transforms.ToTensor())"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "TKaD4RHrxzzO",
        "outputId": "d96ce23e-efa3-431d-8882-c0241c21a131"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Downloading http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz\n",
            "Failed to download (trying next):\n",
            "HTTP Error 403: Forbidden\n",
            "\n",
            "Downloading https://ossci-datasets.s3.amazonaws.com/mnist/train-images-idx3-ubyte.gz\n",
            "Downloading https://ossci-datasets.s3.amazonaws.com/mnist/train-images-idx3-ubyte.gz to data/MNIST/raw/train-images-idx3-ubyte.gz\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "100%|██████████| 9912422/9912422 [00:00<00:00, 156981888.37it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Extracting data/MNIST/raw/train-images-idx3-ubyte.gz to data/MNIST/raw\n",
            "\n",
            "Downloading http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz\n",
            "Failed to download (trying next):\n",
            "HTTP Error 403: Forbidden\n",
            "\n",
            "Downloading https://ossci-datasets.s3.amazonaws.com/mnist/train-labels-idx1-ubyte.gz\n",
            "Downloading https://ossci-datasets.s3.amazonaws.com/mnist/train-labels-idx1-ubyte.gz to data/MNIST/raw/train-labels-idx1-ubyte.gz\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "100%|██████████| 28881/28881 [00:00<00:00, 45031856.44it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Extracting data/MNIST/raw/train-labels-idx1-ubyte.gz to data/MNIST/raw\n",
            "\n",
            "Downloading http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz\n",
            "Failed to download (trying next):\n",
            "HTTP Error 403: Forbidden\n",
            "\n",
            "Downloading https://ossci-datasets.s3.amazonaws.com/mnist/t10k-images-idx3-ubyte.gz\n",
            "Downloading https://ossci-datasets.s3.amazonaws.com/mnist/t10k-images-idx3-ubyte.gz to data/MNIST/raw/t10k-images-idx3-ubyte.gz\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "100%|██████████| 1648877/1648877 [00:00<00:00, 44085925.54it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Extracting data/MNIST/raw/t10k-images-idx3-ubyte.gz to data/MNIST/raw\n",
            "\n",
            "Downloading http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz\n",
            "Failed to download (trying next):\n",
            "HTTP Error 403: Forbidden\n",
            "\n",
            "Downloading https://ossci-datasets.s3.amazonaws.com/mnist/t10k-labels-idx1-ubyte.gz\n",
            "Downloading https://ossci-datasets.s3.amazonaws.com/mnist/t10k-labels-idx1-ubyte.gz to data/MNIST/raw/t10k-labels-idx1-ubyte.gz\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "100%|██████████| 4542/4542 [00:00<00:00, 10636811.15it/s]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Extracting data/MNIST/raw/t10k-labels-idx1-ubyte.gz to data/MNIST/raw\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "dataset"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "1L2YHhoLybbi",
        "outputId": "730f9e02-6c5f-45c6-b8a4-00680397045a"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Dataset MNIST\n",
              "    Number of datapoints: 60000\n",
              "    Root location: data\n",
              "    Split: Train\n",
              "    StandardTransform\n",
              "Transform: ToTensor()"
            ]
          },
          "metadata": {},
          "execution_count": 3
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "generator1 = torch.Generator().manual_seed(42)\n",
        "trainset,testset=torch.utils.data.random_split(dataset,[0.9,0.1],generator=generator1)"
      ],
      "metadata": {
        "id": "5I9NasoFyfIB"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "import matplotlib.pyplot as plt\n",
        "plt.imshow(trainset[0][0].squeeze(0))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 447
        },
        "id": "_eCtFpsKzq4b",
        "outputId": "c6e9f7d7-c8d8-4a5b-a56f-1ac72b6ac3da"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7e0b65a4b460>"
            ]
          },
          "metadata": {},
          "execution_count": 5
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAaAAAAGdCAYAAABU0qcqAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjcuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/bCgiHAAAACXBIWXMAAA9hAAAPYQGoP6dpAAAcJklEQVR4nO3df3DU9b3v8deSH8sPk6UhJJuUgAEFWoE4Uoi5KMWSIaRzHH5N/dk74HVgpMER8NdNR0Xb3qbFU+vVgzBzbgv1jOCPOQIDx9KjgYRrTbCgHC6t5hImllhIUO7NbggmBPK5f3DdupKg32WXd348HzPfGbL7fef74evq0y+7fONzzjkBAHCFDbJeAABgYCJAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACYIEADARLL1Ar6sq6tLx48fV1pamnw+n/VyAAAeOefU2tqq3NxcDRrU83VOrwvQ8ePHlZeXZ70MAMBlamxs1KhRo3p8vtcFKC0tTZJ0k76vZKUYrwYA4NU5deptvRH573lPEhagdevW6emnn1ZTU5MKCgr0/PPPa/r06V859/kfuyUrRck+AgQAfc7/v8PoV72NkpAPIbzyyitavXq11qxZo/fee08FBQUqKSnRyZMnE3E4AEAflJAAPfPMM1q6dKnuueceffvb39aGDRs0dOhQ/fa3v03E4QAAfVDcA3T27FkdOHBAxcXFfz/IoEEqLi5WTU3NRft3dHQoHA5HbQCA/i/uAfr00091/vx5ZWdnRz2enZ2tpqami/avqKhQIBCIbHwCDgAGBvO/iFpeXq5QKBTZGhsbrZcEALgC4v4puMzMTCUlJam5uTnq8ebmZgWDwYv29/v98vv98V4GAKCXi/sVUGpqqqZOnarKysrIY11dXaqsrFRRUVG8DwcA6KMS8veAVq9ercWLF+s73/mOpk+frmeffVZtbW265557EnE4AEAflJAA3X777frkk0/0xBNPqKmpSddff7127dp10QcTAAADl88556wX8UXhcFiBQECzNI87IQBAH3TOdapK2xUKhZSent7jfuafggMADEwECABgggABAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwAQBAgCYIEAAABMECABgItl6AQB6n6Tx4zzPhKdkep75l1/9yvPM6OShnmeSfLH9v/aKvxV6njkyrSOmYw1EXAEBAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACa4GSnQjzVUFMU0d82Nf/U8s2f8CzEcaYjniS457zPuvOeZC8fyxTSHr4crIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACYIEADABDcjBb5g0ODBnmd8qameZz5ZdJ3nmaKy/Z5nXsr+R88zkhQY5P089EdlI/d4nrnzwQc9z+T86h3PM/0BV0AAABMECABgIu4BevLJJ+Xz+aK2iRMnxvswAIA+LiHvAV133XV66623/n6QZN5qAgBES0gZkpOTFQwGE/GtAQD9RELeAzpy5Ihyc3M1duxY3X333Tp27FiP+3Z0dCgcDkdtAID+L+4BKiws1KZNm7Rr1y6tX79eDQ0Nuvnmm9Xa2trt/hUVFQoEApEtLy8v3ksCAPRCcQ9QaWmpfvCDH2jKlCkqKSnRG2+8oZaWFr366qvd7l9eXq5QKBTZGhsb470kAEAvlPBPBwwfPlzjx49XfX19t8/7/X75/f5ELwMA0Msk/O8BnT59WkePHlVOTk6iDwUA6EPiHqCHHnpI1dXV+uijj/TOO+9owYIFSkpK0p133hnvQwEA+rC4/xHcxx9/rDvvvFOnTp3SyJEjddNNN6m2tlYjR46M96EAAH1Y3AP08ssvx/tbAp4lXZMf2+A/d3ge2T5+RwwH8n6Ty9hwU9HLse6TWzzPDNQbi8aCe8EBAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACYS/gPpgMv10X8r8jxz9Y2x/WTdnTHdWBT9Vc3vbvA8kyVuRvp1cQUEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACYIEADABAECAJggQAAAE9wNG5LPF9NYcm6O55m6VaM9z1Td9rTnmeykIZ5ncHkazrV7nln2o5WeZ1JCnZ5nYpX1zr4rdqyBiCsgAIAJAgQAMEGAAAAmCBAAwAQBAgCYIEAAABMECABgggABAEwQIACACQIEADBBgAAAJggQAMAENyOFktLSYprb9u6OOK+kJ9xYNFZz/rIwprnTHX7PM1f9U8DzjH/XnzzPoP/gCggAYIIAAQBMECAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMMHNSAEDDxyf4Xlm77/e4Hlm9IY/e56RpNSWUExzgBdcAQEATBAgAIAJzwHau3evbr31VuXm5srn82nbtm1Rzzvn9MQTTygnJ0dDhgxRcXGxjhw5Eq/1AgD6Cc8BamtrU0FBgdatW9ft82vXrtVzzz2nDRs2aN++fRo2bJhKSkrU3t5+2YsFAPQfnj+EUFpaqtLS0m6fc87p2Wef1WOPPaZ58+ZJkl588UVlZ2dr27ZtuuOOOy5vtQCAfiOu7wE1NDSoqalJxcXFkccCgYAKCwtVU1PT7UxHR4fC4XDUBgDo/+IaoKamJklSdnZ21OPZ2dmR576soqJCgUAgsuXl5cVzSQCAXsr8U3Dl5eUKhUKRrbGx0XpJAIArIK4BCgaDkqTm5uaox5ubmyPPfZnf71d6enrUBgDo/+IaoPz8fAWDQVVWVkYeC4fD2rdvn4qKiuJ5KABAH+f5U3CnT59WfX195OuGhgYdPHhQGRkZGj16tFauXKmf/exnuvbaa5Wfn6/HH39cubm5mj9/fjzXDQDo4zwHaP/+/brlllsiX69evVqStHjxYm3atEmPPPKI2tratGzZMrW0tOimm27Srl27NHjw4PitGgDQ5/mcc856EV8UDocVCAQ0S/OU7EuxXo6pQcOGeZ753z+b7Hnm8dLXPc9I0t1pJ2Ka683+fPac55lXW6Z5nvmPBVd7njn30THPM71d0rVjvQ+FT3seOd980vtxELNzrlNV2q5QKHTJ9/XNPwUHABiYCBAAwAQBAgCYIEAAABMECABgggABAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYMLzj2PAldNaOsnzzIe3rUvASvqefz2dGdPcuv96m+eZtD0fep7525JRnmekWGZ6t22r1nqeub/hB55n/rbtP3mekaQhn3R5ngm8VBvTsQYiroAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwAQBAgCYIEAAABPcjBS93rFzn3meeeER7zeslKTw2CTPM52bR3qe2V/wvOeZ/mmI54mt1+70fpiHvY9I0rsdPs8zDwwp8zwz4n/UeJ7pD7gCAgCYIEAAABMECABgggABAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBM+JxzznoRXxQOhxUIBDRL85TsS7FejqmfN7zreWZKqvebafZ2ne6855k97ekxHevbKZ96nhmV7P2Gmui/Gs61e55Ztnyl5xn/G3/yPHOlnHOdqtJ2hUIhpaf3/O8iV0AAABMECABgggABAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgIlk6wWgZ9enev/H06VedW/ZuEjxeb/B6pwhbTEejRuL4vLkJw/2PHN+8MC8FhiYv2sAgDkCBAAw4TlAe/fu1a233qrc3Fz5fD5t27Yt6vklS5bI5/NFbXPnzo3XegEA/YTnALW1tamgoEDr1q3rcZ+5c+fqxIkTkW3Lli2XtUgAQP/j+V3u0tJSlZaWXnIfv9+vYDAY86IAAP1fQt4DqqqqUlZWliZMmKDly5fr1KlTPe7b0dGhcDgctQEA+r+4B2ju3Ll68cUXVVlZqV/+8peqrq5WaWmpzp8/3+3+FRUVCgQCkS0vLy/eSwIA9EJx/3tAd9xxR+TXkydP1pQpUzRu3DhVVVVp9uzZF+1fXl6u1atXR74Oh8NECAAGgIR/DHvs2LHKzMxUfX19t8/7/X6lp6dHbQCA/i/hAfr444916tQp5eTkJPpQAIA+xPMfwZ0+fTrqaqahoUEHDx5URkaGMjIy9NRTT2nRokUKBoM6evSoHnnkEV1zzTUqKSmJ68IBAH2b5wDt379ft9xyS+Trz9+/Wbx4sdavX69Dhw7pd7/7nVpaWpSbm6s5c+bopz/9qfx+f/xWDQDo8zwHaNasWXKu5xte/uEPf7isBQEABgbuBQcAMEGAAAAmCBAAwAQBAgCYIEAAABMECABgggABAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATcf+R3MBAE+pq9zzz/P+Z7nlmx/qZnmdy/v2E5xlJCl+f7Xkm/+EPPM/8ZvQezzO93R/bUzzPpITPJWAlvR9XQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwAQBAgCYIEAAABMECABgggABAEwQIACACW5G2ovN+I/bPM/8z4JXErASXMrx80meZ17Z7v3Gosr1PvLBg95vKipJdfNfiGkO0n/ZuczzzLVv1SZgJb0fV0AAABMECABgggABAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAluRtqLJf92hPeh/x7/deDSvpWS4nnmf937TwlYCeJtXcs4zzPjX2z1POM8T/QPXAEBAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACa4GWkv5v+/5zzP/PtnwzzPzBnS5nkGsFL52VDPMz9/aHFMxxq265DnGdf+55iONRBxBQQAMEGAAAAmPAWooqJC06ZNU1pamrKysjR//nzV1dVF7dPe3q6ysjKNGDFCV111lRYtWqTm5ua4LhoA0Pd5ClB1dbXKyspUW1urN998U52dnZozZ47a2v7+HsKqVau0Y8cOvfbaa6qurtbx48e1cOHCuC8cANC3efoQwq5du6K+3rRpk7KysnTgwAHNnDlToVBIv/nNb7R582Z973vfkyRt3LhR3/rWt1RbW6sbb7wxfisHAPRpl/UeUCgUkiRlZGRIkg4cOKDOzk4VFxdH9pk4caJGjx6tmpqabr9HR0eHwuFw1AYA6P9iDlBXV5dWrlypGTNmaNKkSZKkpqYmpaamavjw4VH7Zmdnq6mpqdvvU1FRoUAgENny8vJiXRIAoA+JOUBlZWU6fPiwXn755ctaQHl5uUKhUGRrbGy8rO8HAOgbYvqLqCtWrNDOnTu1d+9ejRo1KvJ4MBjU2bNn1dLSEnUV1NzcrGAw2O338vv98vv9sSwDANCHeboCcs5pxYoV2rp1q3bv3q38/Pyo56dOnaqUlBRVVlZGHqurq9OxY8dUVFQUnxUDAPoFT1dAZWVl2rx5s7Zv3660tLTI+zqBQEBDhgxRIBDQvffeq9WrVysjI0Pp6em6//77VVRUxCfgAABRPAVo/fr1kqRZs2ZFPb5x40YtWbJEkvTrX/9agwYN0qJFi9TR0aGSkhK98MILcVksAKD/8DnnnPUivigcDisQCGiW5inZl2K9nD7n9G3erzR//PNNMR2Lm5jico3fsdzzTLDa+2en0rfUep5B7M65TlVpu0KhkNLT03vcj3vBAQBMECAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwERMPxEVvddVr3q/6+/Tp/9zTMea888bYprDlfNvZwIxzf3q0bvjvJLuTfi3g55nXEdH/BcCE1wBAQBMECAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmuBkp5H/jTzHN/cM3p8Z5JegthmrfFTmOuyJHQW/FFRAAwAQBAgCYIEAAABMECABgggABAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwAQBAgCYIEAAABMECABgwlOAKioqNG3aNKWlpSkrK0vz589XXV1d1D6zZs2Sz+eL2u677764LhoA0Pd5ClB1dbXKyspUW1urN998U52dnZozZ47a2tqi9lu6dKlOnDgR2dauXRvXRQMA+r5kLzvv2rUr6utNmzYpKytLBw4c0MyZMyOPDx06VMFgMD4rBAD0S5f1HlAoFJIkZWRkRD3+0ksvKTMzU5MmTVJ5ebnOnDnT4/fo6OhQOByO2gAA/Z+nK6Av6urq0sqVKzVjxgxNmjQp8vhdd92lMWPGKDc3V4cOHdKjjz6quro6vf76691+n4qKCj311FOxLgMA0Ef5nHMulsHly5fr97//vd5++22NGjWqx/12796t2bNnq76+XuPGjbvo+Y6ODnV0dES+DofDysvL0yzNU7IvJZalAQAMnXOdqtJ2hUIhpaen97hfTFdAK1as0M6dO7V3795LxkeSCgsLJanHAPn9fvn9/liWAQDowzwFyDmn+++/X1u3blVVVZXy8/O/cubgwYOSpJycnJgWCADonzwFqKysTJs3b9b27duVlpampqYmSVIgENCQIUN09OhRbd68Wd///vc1YsQIHTp0SKtWrdLMmTM1ZcqUhPwGAAB9k6f3gHw+X7ePb9y4UUuWLFFjY6N++MMf6vDhw2pra1NeXp4WLFigxx577JJ/DvhF4XBYgUCA94AAoI9KyHtAX9WqvLw8VVdXe/mWAIABinvBAQBMECAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwAQBAgCYIEAAABMECABgggABAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMJFsv4Mucc5Kkc+qUnPFiAACenVOnpL//97wnvS5Ara2tkqS39YbxSgAAl6O1tVWBQKDH533uqxJ1hXV1den48eNKS0uTz+eLei4cDisvL0+NjY1KT083WqE9zsMFnIcLOA8XcB4u6A3nwTmn1tZW5ebmatCgnt/p6XVXQIMGDdKoUaMuuU96evqAfoF9jvNwAefhAs7DBZyHC6zPw6WufD7HhxAAACYIEADARJ8KkN/v15o1a+T3+62XYorzcAHn4QLOwwWchwv60nnodR9CAAAMDH3qCggA0H8QIACACQIEADBBgAAAJvpMgNatW6err75agwcPVmFhod59913rJV1xTz75pHw+X9Q2ceJE62Ul3N69e3XrrbcqNzdXPp9P27Zti3reOacnnnhCOTk5GjJkiIqLi3XkyBGbxSbQV52HJUuWXPT6mDt3rs1iE6SiokLTpk1TWlqasrKyNH/+fNXV1UXt097errKyMo0YMUJXXXWVFi1apObmZqMVJ8bXOQ+zZs266PVw3333Ga24e30iQK+88opWr16tNWvW6L333lNBQYFKSkp08uRJ66Vdcdddd51OnDgR2d5++23rJSVcW1ubCgoKtG7dum6fX7t2rZ577jlt2LBB+/bt07Bhw1RSUqL29vYrvNLE+qrzIElz586Nen1s2bLlCq4w8aqrq1VWVqba2lq9+eab6uzs1Jw5c9TW1hbZZ9WqVdqxY4dee+01VVdX6/jx41q4cKHhquPv65wHSVq6dGnU62Ht2rVGK+6B6wOmT5/uysrKIl+fP3/e5ebmuoqKCsNVXXlr1qxxBQUF1sswJclt3bo18nVXV5cLBoPu6aefjjzW0tLi/H6/27Jli8EKr4wvnwfnnFu8eLGbN2+eyXqsnDx50kly1dXVzrkL/+xTUlLca6+9Ftnngw8+cJJcTU2N1TIT7svnwTnnvvvd77oHHnjAblFfQ6+/Ajp79qwOHDig4uLiyGODBg1ScXGxampqDFdm48iRI8rNzdXYsWN1991369ixY9ZLMtXQ0KCmpqao10cgEFBhYeGAfH1UVVUpKytLEyZM0PLly3Xq1CnrJSVUKBSSJGVkZEiSDhw4oM7OzqjXw8SJEzV69Oh+/Xr48nn43EsvvaTMzExNmjRJ5eXlOnPmjMXyetTrbkb6ZZ9++qnOnz+v7OzsqMezs7P14YcfGq3KRmFhoTZt2qQJEyboxIkTeuqpp3TzzTfr8OHDSktLs16eiaamJknq9vXx+XMDxdy5c7Vw4ULl5+fr6NGj+vGPf6zS0lLV1NQoKSnJenlx19XVpZUrV2rGjBmaNGmSpAuvh9TUVA0fPjxq3/78eujuPEjSXXfdpTFjxig3N1eHDh3So48+qrq6Or3++uuGq43W6wOEvystLY38esqUKSosLNSYMWP06quv6t577zVcGXqDO+64I/LryZMna8qUKRo3bpyqqqo0e/Zsw5UlRllZmQ4fPjwg3ge9lJ7Ow7JlyyK/njx5snJycjR79mwdPXpU48aNu9LL7Fav/yO4zMxMJSUlXfQplubmZgWDQaNV9Q7Dhw/X+PHjVV9fb70UM5+/Bnh9XGzs2LHKzMzsl6+PFStWaOfOndqzZ0/Uj28JBoM6e/asWlpaovbvr6+Hns5DdwoLCyWpV70een2AUlNTNXXqVFVWVkYe6+rqUmVlpYqKigxXZu/06dM6evSocnJyrJdiJj8/X8FgMOr1EQ6HtW/fvgH/+vj444916tSpfvX6cM5pxYoV2rp1q3bv3q38/Pyo56dOnaqUlJSo10NdXZ2OHTvWr14PX3UeunPw4EFJ6l2vB+tPQXwdL7/8svP7/W7Tpk3uL3/5i1u2bJkbPny4a2pqsl7aFfXggw+6qqoq19DQ4P74xz+64uJil5mZ6U6ePGm9tIRqbW1177//vnv//fedJPfMM8+4999/3/31r391zjn3i1/8wg0fPtxt377dHTp0yM2bN8/l5+e7zz77zHjl8XWp89Da2uoeeughV1NT4xoaGtxbb73lbrjhBnfttde69vZ266XHzfLly10gEHBVVVXuxIkTke3MmTORfe677z43evRot3v3brd//35XVFTkioqKDFcdf191Hurr691PfvITt3//ftfQ0OC2b9/uxo4d62bOnGm88mh9IkDOOff888+70aNHu9TUVDd9+nRXW1trvaQr7vbbb3c5OTkuNTXVffOb33S33367q6+vt15Wwu3Zs8dJumhbvHixc+7CR7Eff/xxl52d7fx+v5s9e7arq6uzXXQCXOo8nDlzxs2ZM8eNHDnSpaSkuDFjxrilS5f2u/9J6+73L8lt3Lgxss9nn33mfvSjH7lvfOMbbujQoW7BggXuxIkTdotOgK86D8eOHXMzZ850GRkZzu/3u2uuucY9/PDDLhQK2S78S/hxDAAAE73+PSAAQP9EgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACYIEADABAECAJj4fxWrwsats7TIAAAAAElFTkSuQmCC\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "class SparseAE(nn.Module):\n",
        "  def __init__(self):\n",
        "    super().__init__()\n",
        "\n",
        "    self.enc1 = nn.Conv2d(1,8,3,stride=1)\n",
        "    self.enc2 = nn.Conv2d(8,16,5,stride=1)\n",
        "    self.enc3 = nn.Conv2d(16,32,5,stride=2)\n",
        "    self.bottle = nn.MaxPool2d(2, 2)\n",
        "\n",
        "    self.dec1 = nn.ConvTranspose2d(32,16,3, stride=2)\n",
        "    self.dec2 = nn.ConvTranspose2d(16,8,5, stride=2)\n",
        "    self.dec3 = nn.ConvTranspose2d(8,1,8, stride=1)\n",
        "\n",
        "  def encoder(self,x):\n",
        "    x=F.relu(self.enc1(x))\n",
        "    x=F.relu(self.enc2(x))\n",
        "    x=F.relu(self.enc3(x))\n",
        "    x=self.bottle(x)\n",
        "    return x\n",
        "\n",
        "  def decoder(self,x):\n",
        "    x=F.relu(self.dec1(x))\n",
        "    x=F.relu(self.dec2(x))\n",
        "    x=torch.sigmoid(self.dec3(x))\n",
        "    return x\n",
        "\n",
        "\n",
        "  def forward(self,x):\n",
        "    x=self.encoder(x)\n",
        "    x=self.decoder(x)\n",
        "    return x\n",
        "\n",
        "  def forwardwithlayers(self,x):\n",
        "    layers=[]\n",
        "    x1=F.relu(self.enc1(x))\n",
        "    layers.append(x1)\n",
        "    x2=F.relu(self.enc2(x1))\n",
        "    layers.append(x2)\n",
        "    x3=F.relu(self.enc3(x2))\n",
        "    layers.append(x3)\n",
        "    x4=self.bottle(x3)\n",
        "    layers.append(x4)\n",
        "    x5=F.relu(self.dec1(x4))\n",
        "    layers.append(x5)\n",
        "    x6=F.relu(self.dec2(x5))\n",
        "    layers.append(x6)\n",
        "    xans=torch.sigmoid(self.dec3(x6))\n",
        "    return xans,layers\n"
      ],
      "metadata": {
        "id": "1qUhzZ-b0x_j"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def l1_reg(layers, lambda_=4e-5):\n",
        "  loss = 0\n",
        "  for i in layers:\n",
        "    loss += torch.mean(torch.abs(i))\n",
        "\n",
        "  return lambda_*loss"
      ],
      "metadata": {
        "id": "TnB3avMndEQC"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def train(model,train_loader,epochs=3,lr=9e-3,device=\"cpu\"):\n",
        "  model.train()\n",
        "  model.to(device)\n",
        "  optimizer=optim.Adam(model.parameters(),lr=lr)\n",
        "  lossfunc=nn.BCELoss()\n",
        "  lossavg=0\n",
        "\n",
        "  for epoch in range(epochs):\n",
        "    for (bid,(inp,lab)) in enumerate(train_loader):\n",
        "      inp=inp.to(device)\n",
        "      optimizer.zero_grad()\n",
        "      #lab=lab.to(device)\n",
        "      out,layers=model.forwardwithlayers(inp)\n",
        "      loss=lossfunc(out,inp)\n",
        "      loss+=l1_reg(layers)\n",
        "      loss.backward()\n",
        "      optimizer.step()\n",
        "      lossavg+=loss.item()\n",
        "      if bid%10==0:\n",
        "        print(f\"epoch:{epoch} batch={bid} loss:{loss.item()}\")\n",
        "\n",
        "    print(f\"epoch:{epoch} loss:{lossavg/len(train_loader)}\")\n"
      ],
      "metadata": {
        "id": "8KLeE5BtXlKB"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "trainloader=torch.utils.data.DataLoader(trainset,batch_size=128,shuffle=True)\n",
        "testloader=torch.utils.data.DataLoader(testset,batch_size=128,shuffle=False)"
      ],
      "metadata": {
        "id": "seS4GJjeaRKn"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model=SparseAE()\n",
        "device=torch.device(\"cuda\" if torch.cuda.is_available() else 'cpu')\n",
        "train(model,trainloader,epochs=2,device=device)"
      ],
      "metadata": {
        "id": "EygzJcigaWCj"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#test1\n",
        "\n",
        "img=testset[2][0].to(device).unsqueeze(0)\n",
        "out=model(img)\n",
        "plt.imshow(out.detach().cpu().squeeze(0)[0,:,:])\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 447
        },
        "id": "sB02MNY9cI16",
        "outputId": "3027fa12-8e15-4365-bff2-b26125c34118"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7e0b636b95a0>"
            ]
          },
          "metadata": {},
          "execution_count": 32
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plt.imshow(img.detach().cpu().squeeze(0)[0,:,:])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 447
        },
        "id": "L7E8E3Bqe-2K",
        "outputId": "3743d235-ddd8-4291-c0df-9dc2e542e448"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7e0b636ba8c0>"
            ]
          },
          "metadata": {},
          "execution_count": 33
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def test(model,test_loader,device=\"cpu\"):\n",
        "  model.eval()\n",
        "  model.to(device)\n",
        "  lossfunc=nn.BCELoss()\n",
        "  lossavg=0\n",
        "\n",
        "  for (bid,(inp,lab)) in enumerate(test_loader):\n",
        "      inp=inp.to(device)\n",
        "\n",
        "      #lab=lab.to(device)\n",
        "      out,layers=model.forwardwithlayers(inp)\n",
        "      loss=lossfunc(out,inp)\n",
        "      loss+=l1_reg(layers)\n",
        "      lossavg+=loss.item()\n",
        "\n",
        "      print(f\"batch={bid} loss:{loss.item()}\")\n",
        "\n",
        "  print(f\"loss Average:{lossavg/len(test_loader)}\")"
      ],
      "metadata": {
        "id": "WDODHFlAgcS5"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "test(model,testloader,device=device)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "6pITcrWbhlYX",
        "outputId": "f7ff5eaf-50ed-4dbb-85aa-15fce6da6a55"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "batch=0 loss:0.09825733304023743\n",
            "batch=1 loss:0.10166682302951813\n",
            "batch=2 loss:0.10116837918758392\n",
            "batch=3 loss:0.09830772131681442\n",
            "batch=4 loss:0.09583579003810883\n",
            "batch=5 loss:0.0945015549659729\n",
            "batch=6 loss:0.10312231630086899\n",
            "batch=7 loss:0.0961381196975708\n",
            "batch=8 loss:0.10021279752254486\n",
            "batch=9 loss:0.10329338908195496\n",
            "batch=10 loss:0.09609769284725189\n",
            "batch=11 loss:0.09872013330459595\n",
            "batch=12 loss:0.10017648339271545\n",
            "batch=13 loss:0.10135772079229355\n",
            "batch=14 loss:0.10231456160545349\n",
            "batch=15 loss:0.10177243500947952\n",
            "batch=16 loss:0.09996989369392395\n",
            "batch=17 loss:0.0939868837594986\n",
            "batch=18 loss:0.09966843575239182\n",
            "batch=19 loss:0.10404444485902786\n",
            "batch=20 loss:0.09768612682819366\n",
            "batch=21 loss:0.09790600091218948\n",
            "batch=22 loss:0.09703397005796432\n",
            "batch=23 loss:0.09766897559165955\n",
            "batch=24 loss:0.09984725713729858\n",
            "batch=25 loss:0.09305423498153687\n",
            "batch=26 loss:0.09592275321483612\n",
            "batch=27 loss:0.09740911424160004\n",
            "batch=28 loss:0.09581150114536285\n",
            "batch=29 loss:0.0993136391043663\n",
            "batch=30 loss:0.09915225207805634\n",
            "batch=31 loss:0.09894181042909622\n",
            "batch=32 loss:0.10068192332983017\n",
            "batch=33 loss:0.0970747247338295\n",
            "batch=34 loss:0.09607955068349838\n",
            "batch=35 loss:0.10186175256967545\n",
            "batch=36 loss:0.09883149713277817\n",
            "batch=37 loss:0.09909439086914062\n",
            "batch=38 loss:0.09809322655200958\n",
            "batch=39 loss:0.09932908415794373\n",
            "batch=40 loss:0.0976143553853035\n",
            "batch=41 loss:0.09692477434873581\n",
            "batch=42 loss:0.09523144364356995\n",
            "batch=43 loss:0.09944590926170349\n",
            "batch=44 loss:0.0967973917722702\n",
            "batch=45 loss:0.10152988135814667\n",
            "batch=46 loss:0.09702923148870468\n",
            "loss Average:0.09863786557887463\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def kl_reg(layers,rho=0.05,beta=0.005):\n",
        "  loss=0\n",
        "  for i in layers:\n",
        "    loss+=F.kl_div(torch.mean(i,dim=0),rho*torch.ones_like(i[0]))\n",
        "  return beta*torch.abs(loss)"
      ],
      "metadata": {
        "id": "aWrg137_iXx0"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def trainoverKLreg(model,train_loader,epochs=3,lr=9e-3,device=\"cpu\"):\n",
        "  model.train()\n",
        "  model.to(device)\n",
        "  optimizer=optim.Adam(model.parameters(),lr=lr)\n",
        "  lossfunc=nn.BCELoss()\n",
        "  lossavg=0\n",
        "\n",
        "  for epoch in range(epochs):\n",
        "    for (bid,(inp,lab)) in enumerate(train_loader):\n",
        "      inp=inp.to(device)\n",
        "      optimizer.zero_grad()\n",
        "      #lab=lab.to(device)\n",
        "      out,layers=model.forwardwithlayers(inp)\n",
        "      loss=lossfunc(out,inp)\n",
        "      loss+=kl_reg(layers)\n",
        "      loss.backward()\n",
        "      optimizer.step()\n",
        "      lossavg+=loss.item()\n",
        "      if bid%10==0:\n",
        "        print(f\"epoch:{epoch} batch={bid} loss:{loss.item()}\")\n",
        "\n",
        "    print(f\"epoch:{epoch} loss:{lossavg/len(train_loader)}\")\n"
      ],
      "metadata": {
        "id": "-7inxnGgjk_U"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "modelforklreg=SparseAE()\n",
        "device=torch.device(\"cuda\" if torch.cuda.is_available() else 'cpu')\n",
        "#trainoverKLreg(modelforklreg,trainloader,epochs=1,device=device)"
      ],
      "metadata": {
        "id": "esoFgNaqjwsb"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#test1\n",
        "\n",
        "img=testset[2][0].to(device).unsqueeze(0)\n",
        "out=modelforklreg(img)\n",
        "plt.imshow(out.detach().cpu().squeeze(0)[0,:,:])\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 447
        },
        "outputId": "c69f7d10-4a18-4bcd-9544-dd3bf9550959",
        "id": "Ze44uNsNkiwu"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7e0b5ee7c0a0>"
            ]
          },
          "metadata": {},
          "execution_count": 39
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plt.imshow(img.detach().cpu().squeeze(0)[0,:,:])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 447
        },
        "id": "MYlxJEBxkmje",
        "outputId": "7baa6ad3-87ed-49b5-ab82-f66883cc775a"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7e0b5eef03a0>"
            ]
          },
          "metadata": {},
          "execution_count": 40
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAaAAAAGdCAYAAABU0qcqAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjcuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/bCgiHAAAACXBIWXMAAA9hAAAPYQGoP6dpAAAcI0lEQVR4nO3df3DU9b3v8dcSyIqabAwh2aQEDKjQgqS3FNJclaLJJaQzlF+nB390LngdGGjgFtDqTa+K2t5Ji1P16lA8t6cFmRFQ5ghcvZaOBhPGNqEHhEM51ZRw0xIKCZVpsiFICORz/+C67UIifpfdvLPL8zHznSG733e+H77d8dlvdvnG55xzAgCgnw2yXgAA4NpEgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgInB1gu4VE9Pj44fP660tDT5fD7r5QAAPHLOqaOjQ3l5eRo0qO/rnAEXoOPHjys/P996GQCAq9Tc3KwRI0b0+fyAC1BaWpok6U59Q4M1xHg1AACvzqtb7+vt8H/P+xK3AK1du1bPPvusWlpaVFhYqJdeeklTpky54tynP3YbrCEa7CNAAJBw/v8dRq/0NkpcPoTw2muvadWqVVq9erU++OADFRYWqqysTCdPnozH4QAACSguAXruuee0aNEiPfjgg/rSl76kl19+Wddff71+8YtfxONwAIAEFPMAnTt3Tvv27VNpaenfDjJokEpLS1VXV3fZ/l1dXQqFQhEbACD5xTxAH3/8sS5cuKCcnJyIx3NyctTS0nLZ/lVVVQoEAuGNT8ABwLXB/B+iVlZWqr29Pbw1NzdbLwkA0A9i/im4rKwspaSkqLW1NeLx1tZWBYPBy/b3+/3y+/2xXgYAYICL+RVQamqqJk2apOrq6vBjPT09qq6uVnFxcawPBwBIUHH5d0CrVq3SggUL9NWvflVTpkzRCy+8oM7OTj344IPxOBwAIAHFJUDz58/XX/7yFz355JNqaWnRl7/8Ze3cufOyDyYAAK5dPuecs17E3wuFQgoEApqmWdwJAQAS0HnXrRrtUHt7u9LT0/vcz/xTcACAaxMBAgCYIEAAABMECABgggABAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwAQBAgCYIEAAABMECABgggABAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwMdh6AQDiZ/Co/KjmuvMyPc98e/3/8TzzQNpJzzPndcHzzPgtyz3PSNKYh+ujmsPnwxUQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCm5ECSeyj734hqrl/+8cXYruQPpxXiueZbuf9ZqTpt/7V84wkpdw2xvPMhT8ciepY1yKugAAAJggQAMBEzAP01FNPyefzRWzjxo2L9WEAAAkuLu8BjR8/Xu++++7fDjKYt5oAAJHiUobBgwcrGAzG41sDAJJEXN4DOnz4sPLy8jR69Gg98MADOnr0aJ/7dnV1KRQKRWwAgOQX8wAVFRVpw4YN2rlzp9atW6empibddddd6ujo6HX/qqoqBQKB8JafH93vsAcAJJaYB6i8vFzf+ta3NHHiRJWVlentt99WW1ubXn/99V73r6ysVHt7e3hrbm6O9ZIAAANQ3D8dkJGRodtuu02NjY29Pu/3++X3++O9DADAABP3fwd0+vRpHTlyRLm5ufE+FAAggcQ8QI888ohqa2v1xz/+Ub/5zW80Z84cpaSk6L777ov1oQAACSzmP4I7duyY7rvvPp06dUrDhw/XnXfeqfr6eg0fPjzWhwIAJLCYB2jLli2x/pZA0hk8yvunPaO5sWhRUYPnmWS0+yuvRDU3bepKzzPDuBnp58a94AAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACYIEADABAECAJggQAAAE3H/hXQALpeysdvzzL+NeSH2C8FnmrT4gOeZ5trRnmcuHP6/nmeSAVdAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMMHdsIGrdP7dkZ5n/vct/+J5ptt5HsFV+knee55n5t20KA4rSU5cAQEATBAgAIAJAgQAMEGAAAAmCBAAwAQBAgCYIEAAABMECABgggABAEwQIACACQIEADBBgAAAJrgZKQa8lOHDPc989JP8qI71Yck/RTH1r54nBivF+2F83kfGb1nufUjSmIfro5rz6u0/f+B9KIrzgIGJKyAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwAQ3I0W/iubGoh/+j5s9z/zunpc8z0hSt4tqzLsobqi5ozPL80zGhwP7zp3ndcHzTLfzPoOBiSsgAIAJAgQAMOE5QLt379bMmTOVl5cnn8+n7du3RzzvnNOTTz6p3NxcDR06VKWlpTp8+HCs1gsASBKeA9TZ2anCwkKtXbu21+fXrFmjF198US+//LL27NmjG264QWVlZTp79uxVLxYAkDw8fwihvLxc5eXlvT7nnNMLL7ygxx9/XLNmzZIkbdy4UTk5Odq+fbvuvffeq1stACBpxPQ9oKamJrW0tKi0tDT8WCAQUFFRkerq6nqd6erqUigUitgAAMkvpgFqaWmRJOXk5EQ8npOTE37uUlVVVQoEAuEtPz8/lksCAAxQ5p+Cq6ysVHt7e3hrbm62XhIAoB/ENEDBYFCS1NraGvF4a2tr+LlL+f1+paenR2wAgOQX0wAVFBQoGAyquro6/FgoFNKePXtUXFwcy0MBABKc50/BnT59Wo2NjeGvm5qadODAAWVmZmrkyJFasWKFfvjDH+rWW29VQUGBnnjiCeXl5Wn27NmxXDcAIMF5DtDevXt19913h79etWqVJGnBggXasGGDHn30UXV2dmrx4sVqa2vTnXfeqZ07d+q6666L3aoBAAnPc4CmTZsm5/q+Y6PP59MzzzyjZ5555qoWhuT00U+8f8ox2huLDmTjtyz3PBPNjUWH/XPv//wBGAjMPwUHALg2ESAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwITnu2EDn/rDzyZ7nvmo5KeeZ7r7vvl6whrzcL31EgBzXAEBAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACa4GWmSaf7v/9HzzIP3/SqqY/08/fkopvxRHau/TP5fqzzPjNzZEcWRfhfFzMB2/t2RUUz9a8zXgcTBFRAAwAQBAgCYIEAAABMECABgggABAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIKbkQ5gKcOHe545O/as55n/etNHnmcu8n5j0cFK8X4Yn/eR8VuWex+SNOaZ30Q1B2lm7kHPM37fkDisBImCKyAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwAQ3Ix3AWufc4nnm4D3Pe57pdp5HohfFjUV3dGZ5nsn4MIoDIayrfLLnmZtTN3s/juv2PNPtLnieidbKP5d4nkn5a6fnmf77Gw0sXAEBAEwQIACACc8B2r17t2bOnKm8vDz5fD5t37494vmFCxfK5/NFbDNmzIjVegEAScJzgDo7O1VYWKi1a9f2uc+MGTN04sSJ8LZ5s/efDQMAkpvnDyGUl5ervLz8M/fx+/0KBoNRLwoAkPzi8h5QTU2NsrOzNXbsWC1dulSnTp3qc9+uri6FQqGIDQCQ/GIeoBkzZmjjxo2qrq7Wj3/8Y9XW1qq8vFwXLvT+QcOqqioFAoHwlp+fH+slAQAGoJj/O6B77703/Ofbb79dEydO1JgxY1RTU6OSkss/U19ZWalVq1aFvw6FQkQIAK4Bcf8Y9ujRo5WVlaXGxsZen/f7/UpPT4/YAADJL+4BOnbsmE6dOqXc3Nx4HwoAkEA8/wju9OnTEVczTU1NOnDggDIzM5WZmamnn35a8+bNUzAY1JEjR/Too4/qlltuUVlZWUwXDgBIbJ4DtHfvXt19993hrz99/2bBggVat26dDh48qFdeeUVtbW3Ky8vT9OnT9YMf/EB+vz92qwYAJDzPAZo2bZqc6/vulb/61a+uakHJavAo7x+sGPTNvj++nqj2dA3xPLNmzf2eZ4b9c53nmWR0Zk5RVHPf/MG7nmdKh34cxZFSopjpP/t/NtHzzLDDvPY+L+4FBwAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwAQBAgCYIEAAABMx/5Xc6F3Kxm7PM++N2RqHldj6b5VLPM8Me427C0tSV/lkzzPR3NVakpZm/HtUcwPVXfsWRjWXW9vqeeZCVEe6NnEFBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwAQBAgCY4Gak/WTrLW96nul2cViIsbTX6q2XMCCU/3ub55mbUzd7nikd+rHnmWTUcTgjqrnsw7xe44krIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACYIEADABDcj7SevdeR6npmfdiIOK7Hligv75TiNS1Oimvuw5J9ivJLe+X1DPM90ue4ojhTdeegv69pu9TzzP9//T55nbnuEm4oORFwBAQBMECAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmfM45Z72IvxcKhRQIBDRNszQ4ihs2DlTR3ITzja39c2PM/jTE5/3mmN3uQhxWYisZz8O8hn/wPlRyLPYLgbnzrls12qH29nalp6f3uR9XQAAAEwQIAGDCU4Cqqqo0efJkpaWlKTs7W7Nnz1ZDQ0PEPmfPnlVFRYWGDRumG2+8UfPmzVNra2tMFw0ASHyeAlRbW6uKigrV19frnXfeUXd3t6ZPn67Ozs7wPitXrtSbb76prVu3qra2VsePH9fcuXNjvnAAQGLz9BtRd+7cGfH1hg0blJ2drX379mnq1Klqb2/Xz3/+c23atEn33HOPJGn9+vX64he/qPr6en3ta1+L3coBAAntqt4Dam9vlyRlZmZKkvbt26fu7m6VlpaG9xk3bpxGjhypurq6Xr9HV1eXQqFQxAYASH5RB6inp0crVqzQHXfcoQkTJkiSWlpalJqaqoyMjIh9c3Jy1NLS0uv3qaqqUiAQCG/5+fnRLgkAkECiDlBFRYUOHTqkLVu2XNUCKisr1d7eHt6am5uv6vsBABKDp/eAPrVs2TK99dZb2r17t0aMGBF+PBgM6ty5c2pra4u4CmptbVUwGOz1e/n9fvn9/miWAQBIYJ6ugJxzWrZsmbZt26Zdu3apoKAg4vlJkyZpyJAhqq6uDj/W0NCgo0ePqri4ODYrBgAkBU9XQBUVFdq0aZN27NihtLS08Ps6gUBAQ4cOVSAQ0EMPPaRVq1YpMzNT6enpWr58uYqLi/kEHAAggqcArVu3TpI0bdq0iMfXr1+vhQsXSpKef/55DRo0SPPmzVNXV5fKysr005/+NCaLBQAkD25G2k8G54+48k6XOLH2Bs8zu7/yiueZ/pSMN+GMxkA/D3ftW+h5Jq+iw/PM+WN/9jyDgY+bkQIABjQCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwAQBAgCYiOo3osK7883HPM8EHxnjeebrP/nPnmdq/8NGzzPofyv/XOJ5Zv/PJkZ1rNzaVs8z3NkaXnEFBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwAQBAgCY4GakA9iFPxzxPBP8zgjPM/+4ca7nGUl6/dY3opqD9M05/8XzTMpfOz3PDDtc53lGki5ENQV4wxUQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCm5EmmfPNx7wP3R3dsebpa9ENQtLvPE9wg1AkG66AAAAmCBAAwAQBAgCYIEAAABMECABgggABAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAlPAaqqqtLkyZOVlpam7OxszZ49Ww0NDRH7TJs2TT6fL2JbsmRJTBcNAEh8ngJUW1uriooK1dfX65133lF3d7emT5+uzs7OiP0WLVqkEydOhLc1a9bEdNEAgMTn6Tei7ty5M+LrDRs2KDs7W/v27dPUqVPDj19//fUKBoOxWSEAICld1XtA7e3tkqTMzMyIx1999VVlZWVpwoQJqqys1JkzZ/r8Hl1dXQqFQhEbACD5eboC+ns9PT1asWKF7rjjDk2YMCH8+P33369Ro0YpLy9PBw8e1GOPPaaGhga98cYbvX6fqqoqPf3009EuAwCQoHzOORfN4NKlS/XLX/5S77//vkaMGNHnfrt27VJJSYkaGxs1ZsyYy57v6upSV1dX+OtQKKT8/HxN0ywN9g2JZmkAAEPnXbdqtEPt7e1KT0/vc7+oroCWLVumt956S7t37/7M+EhSUVGRJPUZIL/fL7/fH80yAAAJzFOAnHNavny5tm3bppqaGhUUFFxx5sCBA5Kk3NzcqBYIAEhOngJUUVGhTZs2aceOHUpLS1NLS4skKRAIaOjQoTpy5Ig2bdqkb3zjGxo2bJgOHjyolStXaurUqZo4cWJc/gIAgMTk6T0gn8/X6+Pr16/XwoUL1dzcrG9/+9s6dOiQOjs7lZ+frzlz5ujxxx//zJ8D/r1QKKRAIMB7QACQoOLyHtCVWpWfn6/a2lov3xIAcI3iXnAAABMECABgggABAEwQIACACQIEADBBgAAAJggQAMAEAQIAmCBAAAATBAgAYIIAAQBMECAAgAkCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwAQBAgCYIEAAABODrRdwKeecJOm8uiVnvBgAgGfn1S3pb/8978uAC1BHR4ck6X29bbwSAMDV6OjoUCAQ6PN5n7tSovpZT0+Pjh8/rrS0NPl8vojnQqGQ8vPz1dzcrPT0dKMV2uM8XMR5uIjzcBHn4aKBcB6cc+ro6FBeXp4GDer7nZ4BdwU0aNAgjRgx4jP3SU9Pv6ZfYJ/iPFzEebiI83AR5+Ei6/PwWVc+n+JDCAAAEwQIAGAioQLk9/u1evVq+f1+66WY4jxcxHm4iPNwEefhokQ6DwPuQwgAgGtDQl0BAQCSBwECAJggQAAAEwQIAGAiYQK0du1a3XzzzbruuutUVFSk3/72t9ZL6ndPPfWUfD5fxDZu3DjrZcXd7t27NXPmTOXl5cnn82n79u0Rzzvn9OSTTyo3N1dDhw5VaWmpDh8+bLPYOLrSeVi4cOFlr48ZM2bYLDZOqqqqNHnyZKWlpSk7O1uzZ89WQ0NDxD5nz55VRUWFhg0bphtvvFHz5s1Ta2ur0Yrj4/Och2nTpl32eliyZInRinuXEAF67bXXtGrVKq1evVoffPCBCgsLVVZWppMnT1ovrd+NHz9eJ06cCG/vv/++9ZLirrOzU4WFhVq7dm2vz69Zs0YvvviiXn75Ze3Zs0c33HCDysrKdPbs2X5eaXxd6TxI0owZMyJeH5s3b+7HFcZfbW2tKioqVF9fr3feeUfd3d2aPn26Ojs7w/usXLlSb775prZu3ara2lodP35cc+fONVx17H2e8yBJixYting9rFmzxmjFfXAJYMqUKa6ioiL89YULF1xeXp6rqqoyXFX/W716tSssLLRehilJbtu2beGve3p6XDAYdM8++2z4sba2Nuf3+93mzZsNVtg/Lj0Pzjm3YMECN2vWLJP1WDl58qST5Gpra51zF/+3HzJkiNu6dWt4nw8//NBJcnV1dVbLjLtLz4Nzzn3961933/3ud+0W9TkM+Cugc+fOad++fSotLQ0/NmjQIJWWlqqurs5wZTYOHz6svLw8jR49Wg888ICOHj1qvSRTTU1NamlpiXh9BAIBFRUVXZOvj5qaGmVnZ2vs2LFaunSpTp06Zb2kuGpvb5ckZWZmSpL27dun7u7uiNfDuHHjNHLkyKR+PVx6Hj716quvKisrSxMmTFBlZaXOnDljsbw+DbibkV7q448/1oULF5STkxPxeE5Ojj766COjVdkoKirShg0bNHbsWJ04cUJPP/207rrrLh06dEhpaWnWyzPR0tIiSb2+Pj597loxY8YMzZ07VwUFBTpy5Ii+//3vq7y8XHV1dUpJSbFeXsz19PRoxYoVuuOOOzRhwgRJF18PqampysjIiNg3mV8PvZ0HSbr//vs1atQo5eXl6eDBg3rsscfU0NCgN954w3C1kQZ8gPA35eXl4T9PnDhRRUVFGjVqlF5//XU99NBDhivDQHDvvfeG/3z77bdr4sSJGjNmjGpqalRSUmK4svioqKjQoUOHron3QT9LX+dh8eLF4T/ffvvtys3NVUlJiY4cOaIxY8b09zJ7NeB/BJeVlaWUlJTLPsXS2tqqYDBotKqBISMjQ7fddpsaGxutl2Lm09cAr4/LjR49WllZWUn5+li2bJneeustvffeexG/viUYDOrcuXNqa2uL2D9ZXw99nYfeFBUVSdKAej0M+AClpqZq0qRJqq6uDj/W09Oj6upqFRcXG67M3unTp3XkyBHl5uZaL8VMQUGBgsFgxOsjFAppz5491/zr49ixYzp16lRSvT6cc1q2bJm2bdumXbt2qaCgIOL5SZMmaciQIRGvh4aGBh09ejSpXg9XOg+9OXDggCQNrNeD9acgPo8tW7Y4v9/vNmzY4H7/+9+7xYsXu4yMDNfS0mK9tH718MMPu5qaGtfU1OR+/etfu9LSUpeVleVOnjxpvbS46ujocPv373f79+93ktxzzz3n9u/f7/70pz8555z70Y9+5DIyMtyOHTvcwYMH3axZs1xBQYH75JNPjFceW591Hjo6Otwjjzzi6urqXFNTk3v33XfdV77yFXfrrbe6s2fPWi89ZpYuXeoCgYCrqalxJ06cCG9nzpwJ77NkyRI3cuRIt2vXLrd3715XXFzsiouLDVcde1c6D42Nje6ZZ55xe/fudU1NTW7Hjh1u9OjRburUqcYrj5QQAXLOuZdeesmNHDnSpaamuilTprj6+nrrJfW7+fPnu9zcXJeamuq+8IUvuPnz57vGxkbrZcXde++95yRdti1YsMA5d/Gj2E888YTLyclxfr/flZSUuIaGBttFx8FnnYczZ8646dOnu+HDh7shQ4a4UaNGuUWLFiXd/0nr7e8vya1fvz68zyeffOK+853vuJtuusldf/31bs6cOe7EiRN2i46DK52Ho0ePuqlTp7rMzEzn9/vdLbfc4r73ve+59vZ224Vfgl/HAAAwMeDfAwIAJCcCBAAwQYAAACYIEADABAECAJggQAAAEwQIAGCCAAEATBAgAIAJAgQAMEGAAAAmCBAAwMT/A1xM0tKqozptAAAAAElFTkSuQmCC\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def testoverkl(model,test_loader,device=\"cpu\"):\n",
        "  model.eval()\n",
        "  model.to(device)\n",
        "  lossfunc=nn.BCELoss()\n",
        "  lossavg=0\n",
        "\n",
        "  for (bid,(inp,lab)) in enumerate(test_loader):\n",
        "      inp=inp.to(device)\n",
        "\n",
        "      #lab=lab.to(device)\n",
        "      out,layers=model.forwardwithlayers(inp)\n",
        "      loss=lossfunc(out,inp)\n",
        "      loss+=kl_reg(layers)\n",
        "      lossavg+=loss.item()\n",
        "\n",
        "      print(f\"batch={bid} loss:{loss.item()}\")\n",
        "\n",
        "  print(f\"loss Average:{lossavg/len(test_loader)}\")"
      ],
      "metadata": {
        "id": "RDg5Yz0ilJNs"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "test(modelforklreg,testloader,device=device)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "a2EIqNPklT93",
        "outputId": "41c1fc40-371f-4c5a-8163-c2941bea784e"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "batch=0 loss:0.6813063025474548\n",
            "batch=1 loss:0.6809766888618469\n",
            "batch=2 loss:0.6809686422348022\n",
            "batch=3 loss:0.68108731508255\n",
            "batch=4 loss:0.6804786920547485\n",
            "batch=5 loss:0.6807103157043457\n",
            "batch=6 loss:0.6812202334403992\n",
            "batch=7 loss:0.6808823943138123\n",
            "batch=8 loss:0.6806308627128601\n",
            "batch=9 loss:0.6810214519500732\n",
            "batch=10 loss:0.6805139183998108\n",
            "batch=11 loss:0.6806833148002625\n",
            "batch=12 loss:0.6810294389724731\n",
            "batch=13 loss:0.6811744570732117\n",
            "batch=14 loss:0.6809998750686646\n",
            "batch=15 loss:0.6809939742088318\n",
            "batch=16 loss:0.6807049512863159\n",
            "batch=17 loss:0.6805238723754883\n",
            "batch=18 loss:0.6812883019447327\n",
            "batch=19 loss:0.6815454363822937\n",
            "batch=20 loss:0.6806833148002625\n",
            "batch=21 loss:0.680898129940033\n",
            "batch=22 loss:0.6814473271369934\n",
            "batch=23 loss:0.6809667348861694\n",
            "batch=24 loss:0.680898904800415\n",
            "batch=25 loss:0.6804986000061035\n",
            "batch=26 loss:0.6808652281761169\n",
            "batch=27 loss:0.6810388565063477\n",
            "batch=28 loss:0.680804967880249\n",
            "batch=29 loss:0.6813915967941284\n",
            "batch=30 loss:0.6807566285133362\n",
            "batch=31 loss:0.6811339855194092\n",
            "batch=32 loss:0.6811617612838745\n",
            "batch=33 loss:0.6809326410293579\n",
            "batch=34 loss:0.6813051104545593\n",
            "batch=35 loss:0.6811941862106323\n",
            "batch=36 loss:0.681309700012207\n",
            "batch=37 loss:0.6812250018119812\n",
            "batch=38 loss:0.6812711954116821\n",
            "batch=39 loss:0.6810301542282104\n",
            "batch=40 loss:0.6811566352844238\n",
            "batch=41 loss:0.6808743476867676\n",
            "batch=42 loss:0.6810610294342041\n",
            "batch=43 loss:0.6807557344436646\n",
            "batch=44 loss:0.6811569929122925\n",
            "batch=45 loss:0.680802583694458\n",
            "batch=46 loss:0.6808061003684998\n",
            "loss Average:0.6809822955030076\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "torch.save(model.state_dict(),'model.pth')\n",
        "torch.save(modelforklreg.state_dict(),'modelforklreg.pth')"
      ],
      "metadata": {
        "id": "g-tArX3PmM26"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "trainset[0][0].shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "CMMPU45wniUU",
        "outputId": "4b58bd47-1852-4bf4-da80-807e073001c0"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "torch.Size([1, 28, 28])"
            ]
          },
          "metadata": {},
          "execution_count": 44
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "_,layers=model.forwardwithlayers(img)"
      ],
      "metadata": {
        "id": "t7r8Vh8czmCn"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "for i in layers:\n",
        "  print(i.shape)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "32YwKCd-zveu",
        "outputId": "26d37534-34a5-41ed-87e9-084999d53833"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "torch.Size([1, 8, 26, 26])\n",
            "torch.Size([1, 16, 22, 22])\n",
            "torch.Size([1, 32, 9, 9])\n",
            "torch.Size([1, 32, 4, 4])\n",
            "torch.Size([1, 16, 9, 9])\n",
            "torch.Size([1, 8, 21, 21])\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "for param in model.parameters():\n",
        "  param.requires_grad=False"
      ],
      "metadata": {
        "id": "o8jh1fshpVxc"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class labelenc(nn.Module):\n",
        "  def __init__(self):\n",
        "    super().__init__()\n",
        "    self.up = nn.Sequential(\n",
        "            nn.ConvTranspose2d(10, 16, kernel_size=3, stride=2, padding=0), # (16, 2, 2)\n",
        "            nn.ConvTranspose2d(16, 32, kernel_size=2, stride=1, padding=0), # (32, 4, 4)\n",
        "        )\n",
        "  def forward(self,x):\n",
        "\n",
        "    x=self.up(x)\n",
        "    return x"
      ],
      "metadata": {
        "id": "ZPnrB3CHmzF-"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def trainforgen(modellabel,modeldec,train_loader,epochs=3,lr=0.001,device=\"cpu\"):\n",
        "  modellabel.train()\n",
        "  modellabel.to(device)\n",
        "  modeldec.eval()\n",
        "  modeldec.to(device)\n",
        "  optimizer=optim.Adam(modellabel.parameters(),lr=lr)\n",
        "  lossfunc=nn.MSELoss()\n",
        "  lossavg=0\n",
        "\n",
        "  for epoch in range(epochs):\n",
        "    for (bid,(inp,lab)) in enumerate(train_loader):\n",
        "      inp=inp.to(device)\n",
        "      optimizer.zero_grad()\n",
        "      lab=lab.to(device)\n",
        "      batch_size = lab.size(0)\n",
        "      one_hot = torch.zeros(batch_size, 10, 1, 1, dtype=torch.float32, device=device)\n",
        "      one_hot.scatter_(1, lab.view(-1, 1, 1, 1), 1)\n",
        "      #print(one_hot[0,:,:,:])\n",
        "\n",
        "      enc=modellabel(one_hot)\n",
        "      #print(enc.shape)\n",
        "      out=modeldec.encoder(inp)\n",
        "\n",
        "      #print(out.shape)\n",
        "      #print(enc.shape)\n",
        "\n",
        "      loss=lossfunc(enc,out)\n",
        "      loss.backward()\n",
        "      optimizer.step()\n",
        "      lossavg+=loss.item()\n",
        "      if bid%10==0:\n",
        "        print(f\"epoch:{epoch} batch={bid} loss:{loss.item()}\")\n",
        "\n",
        "    print(f\"epoch:{epoch} loss:{lossavg/((epoch+1) * len(train_loader))}\")\n"
      ],
      "metadata": {
        "id": "LyZJz_NQrNR3"
      },
      "execution_count": 110,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "modelforenc=labelenc()"
      ],
      "metadata": {
        "id": "S6FuADm-4yVW"
      },
      "execution_count": 111,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "newtrainloader=torch.utils.data.DataLoader(trainset,batch_size=128,shuffle=True)"
      ],
      "metadata": {
        "id": "WqZRmq1nTf3n"
      },
      "execution_count": 123,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "device=torch.device(\"cuda\" if torch.cuda.is_available() else 'cpu')\n",
        "trainforgen(modelforenc,model,newtrainloader,epochs=10,device=device)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Wgzg0WlMsVYz",
        "outputId": "8b473b2a-d2b4-472e-df8a-141266014340"
      },
      "execution_count": 124,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch:0 batch=0 loss:0.0436556413769722\n",
            "epoch:0 batch=10 loss:0.044273313134908676\n",
            "epoch:0 batch=20 loss:0.046322617679834366\n",
            "epoch:0 batch=30 loss:0.040897730737924576\n",
            "epoch:0 batch=40 loss:0.042505476623773575\n",
            "epoch:0 batch=50 loss:0.04551776126027107\n",
            "epoch:0 batch=60 loss:0.044255297631025314\n",
            "epoch:0 batch=70 loss:0.04356555640697479\n",
            "epoch:0 batch=80 loss:0.04245081543922424\n",
            "epoch:0 batch=90 loss:0.042371805757284164\n",
            "epoch:0 batch=100 loss:0.0413791760802269\n",
            "epoch:0 batch=110 loss:0.04265811666846275\n",
            "epoch:0 batch=120 loss:0.040628258138895035\n",
            "epoch:0 batch=130 loss:0.04262005165219307\n",
            "epoch:0 batch=140 loss:0.04263034462928772\n",
            "epoch:0 batch=150 loss:0.04282480478286743\n",
            "epoch:0 batch=160 loss:0.04081512987613678\n",
            "epoch:0 batch=170 loss:0.04187946021556854\n",
            "epoch:0 batch=180 loss:0.04205167666077614\n",
            "epoch:0 batch=190 loss:0.03985815867781639\n",
            "epoch:0 batch=200 loss:0.037102676928043365\n",
            "epoch:0 batch=210 loss:0.040814246982336044\n",
            "epoch:0 batch=220 loss:0.04085560515522957\n",
            "epoch:0 batch=230 loss:0.03946399688720703\n",
            "epoch:0 batch=240 loss:0.04153992235660553\n",
            "epoch:0 batch=250 loss:0.041496895253658295\n",
            "epoch:0 batch=260 loss:0.04096284508705139\n",
            "epoch:0 batch=270 loss:0.040129538625478745\n",
            "epoch:0 batch=280 loss:0.04054884985089302\n",
            "epoch:0 batch=290 loss:0.040406644344329834\n",
            "epoch:0 batch=300 loss:0.04061121866106987\n",
            "epoch:0 batch=310 loss:0.038121018558740616\n",
            "epoch:0 batch=320 loss:0.03744182735681534\n",
            "epoch:0 batch=330 loss:0.03824155777692795\n",
            "epoch:0 batch=340 loss:0.039896607398986816\n",
            "epoch:0 batch=350 loss:0.0400567427277565\n",
            "epoch:0 batch=360 loss:0.040402963757514954\n",
            "epoch:0 batch=370 loss:0.041391775012016296\n",
            "epoch:0 batch=380 loss:0.04029585421085358\n",
            "epoch:0 batch=390 loss:0.03744414821267128\n",
            "epoch:0 batch=400 loss:0.04033680260181427\n",
            "epoch:0 batch=410 loss:0.04073208197951317\n",
            "epoch:0 batch=420 loss:0.04175906628370285\n",
            "epoch:0 loss:0.04139314679278864\n",
            "epoch:1 batch=0 loss:0.03903721645474434\n",
            "epoch:1 batch=10 loss:0.04217665642499924\n",
            "epoch:1 batch=20 loss:0.042844418436288834\n",
            "epoch:1 batch=30 loss:0.03835982084274292\n",
            "epoch:1 batch=40 loss:0.039766114205121994\n",
            "epoch:1 batch=50 loss:0.0413530096411705\n",
            "epoch:1 batch=60 loss:0.0389690101146698\n",
            "epoch:1 batch=70 loss:0.03975549340248108\n",
            "epoch:1 batch=80 loss:0.038563914597034454\n",
            "epoch:1 batch=90 loss:0.03923472762107849\n",
            "epoch:1 batch=100 loss:0.04004108905792236\n",
            "epoch:1 batch=110 loss:0.04131196439266205\n",
            "epoch:1 batch=120 loss:0.03927510231733322\n",
            "epoch:1 batch=130 loss:0.04334191977977753\n",
            "epoch:1 batch=140 loss:0.04026666656136513\n",
            "epoch:1 batch=150 loss:0.039007995277643204\n",
            "epoch:1 batch=160 loss:0.040329135954380035\n",
            "epoch:1 batch=170 loss:0.038860343396663666\n",
            "epoch:1 batch=180 loss:0.04225882515311241\n",
            "epoch:1 batch=190 loss:0.04336760938167572\n",
            "epoch:1 batch=200 loss:0.04102560505270958\n",
            "epoch:1 batch=210 loss:0.042401496320962906\n",
            "epoch:1 batch=220 loss:0.04028597101569176\n",
            "epoch:1 batch=230 loss:0.04012121632695198\n",
            "epoch:1 batch=240 loss:0.036930713802576065\n",
            "epoch:1 batch=250 loss:0.039992015808820724\n",
            "epoch:1 batch=260 loss:0.04219893366098404\n",
            "epoch:1 batch=270 loss:0.04000823572278023\n",
            "epoch:1 batch=280 loss:0.03889143466949463\n",
            "epoch:1 batch=290 loss:0.039199844002723694\n",
            "epoch:1 batch=300 loss:0.041650399565696716\n",
            "epoch:1 batch=310 loss:0.040346600115299225\n",
            "epoch:1 batch=320 loss:0.040202051401138306\n",
            "epoch:1 batch=330 loss:0.040113769471645355\n",
            "epoch:1 batch=340 loss:0.04158308357000351\n",
            "epoch:1 batch=350 loss:0.041343577206134796\n",
            "epoch:1 batch=360 loss:0.038819875568151474\n",
            "epoch:1 batch=370 loss:0.04049321636557579\n",
            "epoch:1 batch=380 loss:0.04132601246237755\n",
            "epoch:1 batch=390 loss:0.041696444153785706\n",
            "epoch:1 batch=400 loss:0.04010644927620888\n",
            "epoch:1 batch=410 loss:0.041280172765254974\n",
            "epoch:1 batch=420 loss:0.04047093167901039\n",
            "epoch:1 loss:0.04071003940592021\n",
            "epoch:2 batch=0 loss:0.0408344641327858\n",
            "epoch:2 batch=10 loss:0.038878049701452255\n",
            "epoch:2 batch=20 loss:0.04174533113837242\n",
            "epoch:2 batch=30 loss:0.03918693959712982\n",
            "epoch:2 batch=40 loss:0.039908889681100845\n",
            "epoch:2 batch=50 loss:0.04046929255127907\n",
            "epoch:2 batch=60 loss:0.0405045785009861\n",
            "epoch:2 batch=70 loss:0.040067240595817566\n",
            "epoch:2 batch=80 loss:0.03976800665259361\n",
            "epoch:2 batch=90 loss:0.040208231657743454\n",
            "epoch:2 batch=100 loss:0.040688950568437576\n",
            "epoch:2 batch=110 loss:0.036185164004564285\n",
            "epoch:2 batch=120 loss:0.04112982377409935\n",
            "epoch:2 batch=130 loss:0.03945064917206764\n",
            "epoch:2 batch=140 loss:0.03997604548931122\n",
            "epoch:2 batch=150 loss:0.03776004910469055\n",
            "epoch:2 batch=160 loss:0.04177498444914818\n",
            "epoch:2 batch=170 loss:0.040143419057130814\n",
            "epoch:2 batch=180 loss:0.038540828973054886\n",
            "epoch:2 batch=190 loss:0.04031629115343094\n",
            "epoch:2 batch=200 loss:0.03744500130414963\n",
            "epoch:2 batch=210 loss:0.04079924896359444\n",
            "epoch:2 batch=220 loss:0.03709529712796211\n",
            "epoch:2 batch=230 loss:0.03870147466659546\n",
            "epoch:2 batch=240 loss:0.04272161051630974\n",
            "epoch:2 batch=250 loss:0.03910060226917267\n",
            "epoch:2 batch=260 loss:0.04233580827713013\n",
            "epoch:2 batch=270 loss:0.03876563906669617\n",
            "epoch:2 batch=280 loss:0.04081021249294281\n",
            "epoch:2 batch=290 loss:0.037798892706632614\n",
            "epoch:2 batch=300 loss:0.036529120057821274\n",
            "epoch:2 batch=310 loss:0.04042976722121239\n",
            "epoch:2 batch=320 loss:0.04088159650564194\n",
            "epoch:2 batch=330 loss:0.039366647601127625\n",
            "epoch:2 batch=340 loss:0.03810803219676018\n",
            "epoch:2 batch=350 loss:0.03903673589229584\n",
            "epoch:2 batch=360 loss:0.040017738938331604\n",
            "epoch:2 batch=370 loss:0.03779635578393936\n",
            "epoch:2 batch=380 loss:0.03795043006539345\n",
            "epoch:2 batch=390 loss:0.03881271183490753\n",
            "epoch:2 batch=400 loss:0.04182460904121399\n",
            "epoch:2 batch=410 loss:0.039703041315078735\n",
            "epoch:2 batch=420 loss:0.040888573974370956\n",
            "epoch:2 loss:0.040431118292195534\n",
            "epoch:3 batch=0 loss:0.04088284447789192\n",
            "epoch:3 batch=10 loss:0.03930843621492386\n",
            "epoch:3 batch=20 loss:0.0379105806350708\n",
            "epoch:3 batch=30 loss:0.04078381136059761\n",
            "epoch:3 batch=40 loss:0.04260784387588501\n",
            "epoch:3 batch=50 loss:0.03795431926846504\n",
            "epoch:3 batch=60 loss:0.03929227590560913\n",
            "epoch:3 batch=70 loss:0.04088965803384781\n",
            "epoch:3 batch=80 loss:0.03869006782770157\n",
            "epoch:3 batch=90 loss:0.03968867287039757\n",
            "epoch:3 batch=100 loss:0.0386686697602272\n",
            "epoch:3 batch=110 loss:0.04087807983160019\n",
            "epoch:3 batch=120 loss:0.03965499624609947\n",
            "epoch:3 batch=130 loss:0.03895598277449608\n",
            "epoch:3 batch=140 loss:0.03944513574242592\n",
            "epoch:3 batch=150 loss:0.040245965123176575\n",
            "epoch:3 batch=160 loss:0.03963674604892731\n",
            "epoch:3 batch=170 loss:0.03924534097313881\n",
            "epoch:3 batch=180 loss:0.039262205362319946\n",
            "epoch:3 batch=190 loss:0.037823572754859924\n",
            "epoch:3 batch=200 loss:0.037330567836761475\n",
            "epoch:3 batch=210 loss:0.039352718740701675\n",
            "epoch:3 batch=220 loss:0.04207349568605423\n",
            "epoch:3 batch=230 loss:0.04019470512866974\n",
            "epoch:3 batch=240 loss:0.037840429693460464\n",
            "epoch:3 batch=250 loss:0.04047388210892677\n",
            "epoch:3 batch=260 loss:0.03985769301652908\n",
            "epoch:3 batch=270 loss:0.03929855301976204\n",
            "epoch:3 batch=280 loss:0.038564667105674744\n",
            "epoch:3 batch=290 loss:0.039790913462638855\n",
            "epoch:3 batch=300 loss:0.040217168629169464\n",
            "epoch:3 batch=310 loss:0.041071370244026184\n",
            "epoch:3 batch=320 loss:0.041174352169036865\n",
            "epoch:3 batch=330 loss:0.03727448731660843\n",
            "epoch:3 batch=340 loss:0.040767136961221695\n",
            "epoch:3 batch=350 loss:0.04126109182834625\n",
            "epoch:3 batch=360 loss:0.03826676309108734\n",
            "epoch:3 batch=370 loss:0.03714975714683533\n",
            "epoch:3 batch=380 loss:0.04023830592632294\n",
            "epoch:3 batch=390 loss:0.03956630825996399\n",
            "epoch:3 batch=400 loss:0.04257858917117119\n",
            "epoch:3 batch=410 loss:0.03556925430893898\n",
            "epoch:3 batch=420 loss:0.04097352549433708\n",
            "epoch:3 loss:0.040280579779102904\n",
            "epoch:4 batch=0 loss:0.03898371756076813\n",
            "epoch:4 batch=10 loss:0.0402294285595417\n",
            "epoch:4 batch=20 loss:0.04082479327917099\n",
            "epoch:4 batch=30 loss:0.041058268398046494\n",
            "epoch:4 batch=40 loss:0.03851712867617607\n",
            "epoch:4 batch=50 loss:0.03935351222753525\n",
            "epoch:4 batch=60 loss:0.039350174367427826\n",
            "epoch:4 batch=70 loss:0.03863745927810669\n",
            "epoch:4 batch=80 loss:0.04037129506468773\n",
            "epoch:4 batch=90 loss:0.043753113597631454\n",
            "epoch:4 batch=100 loss:0.03862737864255905\n",
            "epoch:4 batch=110 loss:0.039969101548194885\n",
            "epoch:4 batch=120 loss:0.03850486874580383\n",
            "epoch:4 batch=130 loss:0.03905618190765381\n",
            "epoch:4 batch=140 loss:0.03922363370656967\n",
            "epoch:4 batch=150 loss:0.03825884312391281\n",
            "epoch:4 batch=160 loss:0.038544438779354095\n",
            "epoch:4 batch=170 loss:0.03715500608086586\n",
            "epoch:4 batch=180 loss:0.040107693523168564\n",
            "epoch:4 batch=190 loss:0.04023535177111626\n",
            "epoch:4 batch=200 loss:0.04087228700518608\n",
            "epoch:4 batch=210 loss:0.03832981362938881\n",
            "epoch:4 batch=220 loss:0.04162762686610222\n",
            "epoch:4 batch=230 loss:0.04148475080728531\n",
            "epoch:4 batch=240 loss:0.04000948742032051\n",
            "epoch:4 batch=250 loss:0.03946882486343384\n",
            "epoch:4 batch=260 loss:0.03959772363305092\n",
            "epoch:4 batch=270 loss:0.03949084132909775\n",
            "epoch:4 batch=280 loss:0.03958730399608612\n",
            "epoch:4 batch=290 loss:0.040952734649181366\n",
            "epoch:4 batch=300 loss:0.0410701148211956\n",
            "epoch:4 batch=310 loss:0.038086242973804474\n",
            "epoch:4 batch=320 loss:0.03859345614910126\n",
            "epoch:4 batch=330 loss:0.03901754692196846\n",
            "epoch:4 batch=340 loss:0.042407646775245667\n",
            "epoch:4 batch=350 loss:0.040860675275325775\n",
            "epoch:4 batch=360 loss:0.039443667978048325\n",
            "epoch:4 batch=370 loss:0.03673519194126129\n",
            "epoch:4 batch=380 loss:0.03809737414121628\n",
            "epoch:4 batch=390 loss:0.03894086927175522\n",
            "epoch:4 batch=400 loss:0.04021536186337471\n",
            "epoch:4 batch=410 loss:0.04066783934831619\n",
            "epoch:4 batch=420 loss:0.04052218794822693\n",
            "epoch:4 loss:0.04018637907660403\n",
            "epoch:5 batch=0 loss:0.042078036814928055\n",
            "epoch:5 batch=10 loss:0.037932898849248886\n",
            "epoch:5 batch=20 loss:0.0364854596555233\n",
            "epoch:5 batch=30 loss:0.0367271713912487\n",
            "epoch:5 batch=40 loss:0.04003344848752022\n",
            "epoch:5 batch=50 loss:0.037470459938049316\n",
            "epoch:5 batch=60 loss:0.03970414400100708\n",
            "epoch:5 batch=70 loss:0.038858186453580856\n",
            "epoch:5 batch=80 loss:0.039648693054914474\n",
            "epoch:5 batch=90 loss:0.04224688559770584\n",
            "epoch:5 batch=100 loss:0.037708181887865067\n",
            "epoch:5 batch=110 loss:0.03935376554727554\n",
            "epoch:5 batch=120 loss:0.040065545588731766\n",
            "epoch:5 batch=130 loss:0.03671957552433014\n",
            "epoch:5 batch=140 loss:0.038410965353250504\n",
            "epoch:5 batch=150 loss:0.037479110062122345\n",
            "epoch:5 batch=160 loss:0.04014094918966293\n",
            "epoch:5 batch=170 loss:0.038333166390657425\n",
            "epoch:5 batch=180 loss:0.03980211913585663\n",
            "epoch:5 batch=190 loss:0.03958728909492493\n",
            "epoch:5 batch=200 loss:0.03951193392276764\n",
            "epoch:5 batch=210 loss:0.04158427193760872\n",
            "epoch:5 batch=220 loss:0.0401473306119442\n",
            "epoch:5 batch=230 loss:0.03790327534079552\n",
            "epoch:5 batch=240 loss:0.040094856172800064\n",
            "epoch:5 batch=250 loss:0.040312036871910095\n",
            "epoch:5 batch=260 loss:0.036898110061883926\n",
            "epoch:5 batch=270 loss:0.04058614373207092\n",
            "epoch:5 batch=280 loss:0.038950223475694656\n",
            "epoch:5 batch=290 loss:0.04033639654517174\n",
            "epoch:5 batch=300 loss:0.04081322252750397\n",
            "epoch:5 batch=310 loss:0.038944847881793976\n",
            "epoch:5 batch=320 loss:0.04068242013454437\n",
            "epoch:5 batch=330 loss:0.03941415250301361\n",
            "epoch:5 batch=340 loss:0.03955478221178055\n",
            "epoch:5 batch=350 loss:0.035868749022483826\n",
            "epoch:5 batch=360 loss:0.037603192031383514\n",
            "epoch:5 batch=370 loss:0.040987417101860046\n",
            "epoch:5 batch=380 loss:0.04107002913951874\n",
            "epoch:5 batch=390 loss:0.03952447697520256\n",
            "epoch:5 batch=400 loss:0.042972613126039505\n",
            "epoch:5 batch=410 loss:0.04087819159030914\n",
            "epoch:5 batch=420 loss:0.0396997332572937\n",
            "epoch:5 loss:0.040120918411131815\n",
            "epoch:6 batch=0 loss:0.040451448410749435\n",
            "epoch:6 batch=10 loss:0.03937377408146858\n",
            "epoch:6 batch=20 loss:0.0393318310379982\n",
            "epoch:6 batch=30 loss:0.03910032659769058\n",
            "epoch:6 batch=40 loss:0.04169083386659622\n",
            "epoch:6 batch=50 loss:0.039076317101716995\n",
            "epoch:6 batch=60 loss:0.03722897171974182\n",
            "epoch:6 batch=70 loss:0.03813544288277626\n",
            "epoch:6 batch=80 loss:0.03824504092335701\n",
            "epoch:6 batch=90 loss:0.03969895839691162\n",
            "epoch:6 batch=100 loss:0.039586883038282394\n",
            "epoch:6 batch=110 loss:0.04177992045879364\n",
            "epoch:6 batch=120 loss:0.04121968150138855\n",
            "epoch:6 batch=130 loss:0.036309678107500076\n",
            "epoch:6 batch=140 loss:0.03668108582496643\n",
            "epoch:6 batch=150 loss:0.041172198951244354\n",
            "epoch:6 batch=160 loss:0.042506393045186996\n",
            "epoch:6 batch=170 loss:0.03812131658196449\n",
            "epoch:6 batch=180 loss:0.03865356743335724\n",
            "epoch:6 batch=190 loss:0.0382375530898571\n",
            "epoch:6 batch=200 loss:0.0396575964987278\n",
            "epoch:6 batch=210 loss:0.039607204496860504\n",
            "epoch:6 batch=220 loss:0.04103923961520195\n",
            "epoch:6 batch=230 loss:0.04283739626407623\n",
            "epoch:6 batch=240 loss:0.0369754359126091\n",
            "epoch:6 batch=250 loss:0.04005300626158714\n",
            "epoch:6 batch=260 loss:0.040417563170194626\n",
            "epoch:6 batch=270 loss:0.03866682946681976\n",
            "epoch:6 batch=280 loss:0.03884711489081383\n",
            "epoch:6 batch=290 loss:0.03776058927178383\n",
            "epoch:6 batch=300 loss:0.040857914835214615\n",
            "epoch:6 batch=310 loss:0.03937786817550659\n",
            "epoch:6 batch=320 loss:0.040386419743299484\n",
            "epoch:6 batch=330 loss:0.04226406291127205\n",
            "epoch:6 batch=340 loss:0.0400049164891243\n",
            "epoch:6 batch=350 loss:0.041744157671928406\n",
            "epoch:6 batch=360 loss:0.03999903053045273\n",
            "epoch:6 batch=370 loss:0.04106505587697029\n",
            "epoch:6 batch=380 loss:0.03843251243233681\n",
            "epoch:6 batch=390 loss:0.03667566552758217\n",
            "epoch:6 batch=400 loss:0.0384671613574028\n",
            "epoch:6 batch=410 loss:0.04143190011382103\n",
            "epoch:6 batch=420 loss:0.03932640701532364\n",
            "epoch:6 loss:0.04007350169904101\n",
            "epoch:7 batch=0 loss:0.04050742834806442\n",
            "epoch:7 batch=10 loss:0.04113622382283211\n",
            "epoch:7 batch=20 loss:0.039960917085409164\n",
            "epoch:7 batch=30 loss:0.039709463715553284\n",
            "epoch:7 batch=40 loss:0.038390133529901505\n",
            "epoch:7 batch=50 loss:0.03890639543533325\n",
            "epoch:7 batch=60 loss:0.041833773255348206\n",
            "epoch:7 batch=70 loss:0.03905152529478073\n",
            "epoch:7 batch=80 loss:0.03966803103685379\n",
            "epoch:7 batch=90 loss:0.0401829332113266\n",
            "epoch:7 batch=100 loss:0.04177815094590187\n",
            "epoch:7 batch=110 loss:0.03969429060816765\n",
            "epoch:7 batch=120 loss:0.03989259526133537\n",
            "epoch:7 batch=130 loss:0.038340844213962555\n",
            "epoch:7 batch=140 loss:0.037682436406612396\n",
            "epoch:7 batch=150 loss:0.04196137562394142\n",
            "epoch:7 batch=160 loss:0.040775198489427567\n",
            "epoch:7 batch=170 loss:0.03954349830746651\n",
            "epoch:7 batch=180 loss:0.04099811613559723\n",
            "epoch:7 batch=190 loss:0.040102846920490265\n",
            "epoch:7 batch=200 loss:0.042688123881816864\n",
            "epoch:7 batch=210 loss:0.03876952826976776\n",
            "epoch:7 batch=220 loss:0.037972141057252884\n",
            "epoch:7 batch=230 loss:0.03716868534684181\n",
            "epoch:7 batch=240 loss:0.038962289690971375\n",
            "epoch:7 batch=250 loss:0.04121159762144089\n",
            "epoch:7 batch=260 loss:0.03730418533086777\n",
            "epoch:7 batch=270 loss:0.03971821814775467\n",
            "epoch:7 batch=280 loss:0.037703923881053925\n",
            "epoch:7 batch=290 loss:0.039180174469947815\n",
            "epoch:7 batch=300 loss:0.041005685925483704\n",
            "epoch:7 batch=310 loss:0.03965145722031593\n",
            "epoch:7 batch=320 loss:0.03889467939734459\n",
            "epoch:7 batch=330 loss:0.04119497537612915\n",
            "epoch:7 batch=340 loss:0.03616584464907646\n",
            "epoch:7 batch=350 loss:0.03923984616994858\n",
            "epoch:7 batch=360 loss:0.03852718323469162\n",
            "epoch:7 batch=370 loss:0.04019811749458313\n",
            "epoch:7 batch=380 loss:0.04195536673069\n",
            "epoch:7 batch=390 loss:0.03951619192957878\n",
            "epoch:7 batch=400 loss:0.0379018560051918\n",
            "epoch:7 batch=410 loss:0.03906090557575226\n",
            "epoch:7 batch=420 loss:0.04012467712163925\n",
            "epoch:7 loss:0.04003719073583447\n",
            "epoch:8 batch=0 loss:0.041363004595041275\n",
            "epoch:8 batch=10 loss:0.04035525768995285\n",
            "epoch:8 batch=20 loss:0.04474884644150734\n",
            "epoch:8 batch=30 loss:0.03707944229245186\n",
            "epoch:8 batch=40 loss:0.03988298028707504\n",
            "epoch:8 batch=50 loss:0.04166029393672943\n",
            "epoch:8 batch=60 loss:0.036823779344558716\n",
            "epoch:8 batch=70 loss:0.04264364764094353\n",
            "epoch:8 batch=80 loss:0.03690601885318756\n",
            "epoch:8 batch=90 loss:0.037417612969875336\n",
            "epoch:8 batch=100 loss:0.03844960033893585\n",
            "epoch:8 batch=110 loss:0.03887761011719704\n",
            "epoch:8 batch=120 loss:0.03773710876703262\n",
            "epoch:8 batch=130 loss:0.03980351611971855\n",
            "epoch:8 batch=140 loss:0.03919127210974693\n",
            "epoch:8 batch=150 loss:0.039662715047597885\n",
            "epoch:8 batch=160 loss:0.03831186890602112\n",
            "epoch:8 batch=170 loss:0.0403221920132637\n",
            "epoch:8 batch=180 loss:0.042006492614746094\n",
            "epoch:8 batch=190 loss:0.038502130657434464\n",
            "epoch:8 batch=200 loss:0.037402503192424774\n",
            "epoch:8 batch=210 loss:0.04175633192062378\n",
            "epoch:8 batch=220 loss:0.0408528596162796\n",
            "epoch:8 batch=230 loss:0.0405491478741169\n",
            "epoch:8 batch=240 loss:0.03931625932455063\n",
            "epoch:8 batch=250 loss:0.03890557959675789\n",
            "epoch:8 batch=260 loss:0.040962785482406616\n",
            "epoch:8 batch=270 loss:0.04262650012969971\n",
            "epoch:8 batch=280 loss:0.039578262716531754\n",
            "epoch:8 batch=290 loss:0.038143038749694824\n",
            "epoch:8 batch=300 loss:0.039981916546821594\n",
            "epoch:8 batch=310 loss:0.04149084910750389\n",
            "epoch:8 batch=320 loss:0.038771774619817734\n",
            "epoch:8 batch=330 loss:0.04093583673238754\n",
            "epoch:8 batch=340 loss:0.03946860879659653\n",
            "epoch:8 batch=350 loss:0.03940323367714882\n",
            "epoch:8 batch=360 loss:0.0393533930182457\n",
            "epoch:8 batch=370 loss:0.039916347712278366\n",
            "epoch:8 batch=380 loss:0.03866610303521156\n",
            "epoch:8 batch=390 loss:0.040110230445861816\n",
            "epoch:8 batch=400 loss:0.04027421399950981\n",
            "epoch:8 batch=410 loss:0.04076281189918518\n",
            "epoch:8 batch=420 loss:0.038508880883455276\n",
            "epoch:8 loss:0.040009320979882\n",
            "epoch:9 batch=0 loss:0.04094521701335907\n",
            "epoch:9 batch=10 loss:0.03862139210104942\n",
            "epoch:9 batch=20 loss:0.038883160799741745\n",
            "epoch:9 batch=30 loss:0.03824559971690178\n",
            "epoch:9 batch=40 loss:0.04116281121969223\n",
            "epoch:9 batch=50 loss:0.041050150990486145\n",
            "epoch:9 batch=60 loss:0.04100365564227104\n",
            "epoch:9 batch=70 loss:0.04079405590891838\n",
            "epoch:9 batch=80 loss:0.04210568591952324\n",
            "epoch:9 batch=90 loss:0.037976767867803574\n",
            "epoch:9 batch=100 loss:0.03826296702027321\n",
            "epoch:9 batch=110 loss:0.03923925757408142\n",
            "epoch:9 batch=120 loss:0.04120182245969772\n",
            "epoch:9 batch=130 loss:0.04066179692745209\n",
            "epoch:9 batch=140 loss:0.04024209827184677\n",
            "epoch:9 batch=150 loss:0.038960278034210205\n",
            "epoch:9 batch=160 loss:0.04105231910943985\n",
            "epoch:9 batch=170 loss:0.03799373283982277\n",
            "epoch:9 batch=180 loss:0.03871523588895798\n",
            "epoch:9 batch=190 loss:0.039342306554317474\n",
            "epoch:9 batch=200 loss:0.04097382351756096\n",
            "epoch:9 batch=210 loss:0.042608655989170074\n",
            "epoch:9 batch=220 loss:0.040853116661310196\n",
            "epoch:9 batch=230 loss:0.0397607795894146\n",
            "epoch:9 batch=240 loss:0.040634170174598694\n",
            "epoch:9 batch=250 loss:0.04009857028722763\n",
            "epoch:9 batch=260 loss:0.040983304381370544\n",
            "epoch:9 batch=270 loss:0.03991652652621269\n",
            "epoch:9 batch=280 loss:0.039914317429065704\n",
            "epoch:9 batch=290 loss:0.04177234694361687\n",
            "epoch:9 batch=300 loss:0.03749643266201019\n",
            "epoch:9 batch=310 loss:0.04154049605131149\n",
            "epoch:9 batch=320 loss:0.040096160024404526\n",
            "epoch:9 batch=330 loss:0.039427656680345535\n",
            "epoch:9 batch=340 loss:0.04228188842535019\n",
            "epoch:9 batch=350 loss:0.03931182622909546\n",
            "epoch:9 batch=360 loss:0.040667690336704254\n",
            "epoch:9 batch=370 loss:0.039830245077610016\n",
            "epoch:9 batch=380 loss:0.040338534861803055\n",
            "epoch:9 batch=390 loss:0.042878806591033936\n",
            "epoch:9 batch=400 loss:0.03637537360191345\n",
            "epoch:9 batch=410 loss:0.03680633753538132\n",
            "epoch:9 batch=420 loss:0.036223508417606354\n",
            "epoch:9 loss:0.0399862486595438\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "torch.save(modelforenc.state_dict(),'modelforencfinal.pth')"
      ],
      "metadata": {
        "id": "qrgjtIc2wHNw"
      },
      "execution_count": 125,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def runr(lab,modelforencc=modelforenc,modelfordecc=model):\n",
        "  x=torch.zeros(10)\n",
        "  x[lab]=1\n",
        "  x=x.view(10,1,1)\n",
        "  x=x.to(device)\n",
        "  x=x.unsqueeze(0)\n",
        "  #print(x)\n",
        "\n",
        "  modelforencc.eval()\n",
        "  modelfordecc.eval()\n",
        "\n",
        "  enc=modelforencc(x)\n",
        "  out=modelfordecc.decoder(enc)\n",
        "  return out\n"
      ],
      "metadata": {
        "id": "ibXT4FuE2tDV"
      },
      "execution_count": 126,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "out=runr(0)\n",
        "plt.imshow(out.detach().cpu().squeeze(0,1))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 447
        },
        "id": "gHS-_6z14dJ_",
        "outputId": "c1405ca9-e747-4157-a29e-0071c3115b49"
      },
      "execution_count": 129,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7e0b5cf4b250>"
            ]
          },
          "metadata": {},
          "execution_count": 129
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plt.figure(figsize=(20, 4))\n",
        "for i in range(10):\n",
        "    ax = plt.subplot(1, 10, i+1)\n",
        "    out=runr(i)\n",
        "    plt.imshow(out.detach().cpu().squeeze(0,1),cmap='gray')\n",
        "    ax.set_title(f\"Label: {i}\")\n",
        "    plt.axis('off')\n",
        "plt.tight_layout()\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 122
        },
        "id": "-i9GkYrKbGct",
        "outputId": "4bef2006-3806-4d49-f89f-97e3034d5137"
      },
      "execution_count": 132,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 2000x400 with 10 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    }
  ]
}