Preetam_Saha_AutoEncoder_CNN_03_09_2022.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "collapsed_sections": [],
      "authorship_tag": "ABX9TyPA4PzWXWqWMrZzzHoLVw4K",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/preetamjumech/Autoencoder_cnn-and-VAE/blob/main/Preetam_Saha_AutoEncoder_CNN_03_09_2022.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "id": "o935d1xWAfPc"
      },
      "outputs": [],
      "source": [
        "import torch\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F\n",
        "from torch.utils.data import DataLoader, TensorDataset\n",
        "from torch.optim import Adam \n",
        "import matplotlib.pyplot as plt"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#cnn autoencoder with noise removal"
      ],
      "metadata": {
        "id": "LB1bzcHNtf44"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
        "device"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yULH6RM5BMde",
        "outputId": "40213a7b-41b3-41e8-902c-b1e43dcd7b8d"
      },
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "device(type='cpu')"
            ]
          },
          "metadata": {},
          "execution_count": 2
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "!wget https://github.com/MorvanZhou/PyTorch-Tutorial/raw/master/tutorial-contents-notebooks/mnist/processed/training.pt\n",
        "!wget https://github.com/MorvanZhou/PyTorch-Tutorial/raw/master/tutorial-contents-notebooks/mnist/processed/test.pt"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "r9e8Lmv_BYbs",
        "outputId": "f0754cfa-8cb4-4b72-aa8d-7d88fe69f332"
      },
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "--2022-10-04 07:12:22--  https://github.com/MorvanZhou/PyTorch-Tutorial/raw/master/tutorial-contents-notebooks/mnist/processed/training.pt\n",
            "Resolving github.com (github.com)... 140.82.114.3\n",
            "Connecting to github.com (github.com)|140.82.114.3|:443... connected.\n",
            "HTTP request sent, awaiting response... 302 Found\n",
            "Location: https://raw.githubusercontent.com/MorvanZhou/PyTorch-Tutorial/master/tutorial-contents-notebooks/mnist/processed/training.pt [following]\n",
            "--2022-10-04 07:12:23--  https://raw.githubusercontent.com/MorvanZhou/PyTorch-Tutorial/master/tutorial-contents-notebooks/mnist/processed/training.pt\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.111.133, 185.199.108.133, 185.199.110.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.111.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 47520431 (45M) [application/octet-stream]\n",
            "Saving to: ‘training.pt’\n",
            "\n",
            "training.pt         100%[===================>]  45.32M   193MB/s    in 0.2s    \n",
            "\n",
            "2022-10-04 07:12:23 (193 MB/s) - ‘training.pt’ saved [47520431/47520431]\n",
            "\n",
            "--2022-10-04 07:12:24--  https://github.com/MorvanZhou/PyTorch-Tutorial/raw/master/tutorial-contents-notebooks/mnist/processed/test.pt\n",
            "Resolving github.com (github.com)... 140.82.114.3\n",
            "Connecting to github.com (github.com)|140.82.114.3|:443... connected.\n",
            "HTTP request sent, awaiting response... 302 Found\n",
            "Location: https://raw.githubusercontent.com/MorvanZhou/PyTorch-Tutorial/master/tutorial-contents-notebooks/mnist/processed/test.pt [following]\n",
            "--2022-10-04 07:12:24--  https://raw.githubusercontent.com/MorvanZhou/PyTorch-Tutorial/master/tutorial-contents-notebooks/mnist/processed/test.pt\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.108.133, 185.199.109.133, 185.199.110.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.108.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 7920431 (7.6M) [application/octet-stream]\n",
            "Saving to: ‘test.pt’\n",
            "\n",
            "test.pt             100%[===================>]   7.55M  --.-KB/s    in 0.09s   \n",
            "\n",
            "2022-10-04 07:12:24 (88.4 MB/s) - ‘test.pt’ saved [7920431/7920431]\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "x_train, _ = torch.load(\"training.pt\") #for auto encoder, we dont need y_train, we are using MNIST data, training will be faster"
      ],
      "metadata": {
        "id": "BgsghgKsGTKt"
      },
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "x_train.shape"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "zxLWlJKWKKEM",
        "outputId": "764a49cd-7993-4c32-bff4-913016054a05"
      },
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "torch.Size([60000, 28, 28])"
            ]
          },
          "metadata": {},
          "execution_count": 5
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plt.imshow(x_train[12]) #image without the normalization"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "id": "7dc1vq-9KKRu",
        "outputId": "1944ffc5-85cf-46aa-ffe8-e7edab4122a9"
      },
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f2cbdf92a10>"
            ]
          },
          "metadata": {},
          "execution_count": 6
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def normalize(x):\n",
        "  max_val = x.max()\n",
        "  min_val = x.min()\n",
        "  return (x-min_val)/(max_val - min_val)"
      ],
      "metadata": {
        "id": "qJPQFc-yGb4r"
      },
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#x_train = normalize(x_train)"
      ],
      "metadata": {
        "id": "Q_t8vNvrHMYV"
      },
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "x_train.shape #28*28 is a single image, shape remaining same, only values are getting normalized "
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "f4yl6o7JHQPO",
        "outputId": "6ab20feb-9742-4b9c-e70d-240d36aac28d"
      },
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "torch.Size([60000, 28, 28])"
            ]
          },
          "metadata": {},
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def salt_n_pepper_noise(x):\n",
        "  return x+torch.rand(x.size())\n",
        "\n",
        "def gaussian_noise(x):\n",
        "  return x+torch.normal(0,0.15,size=x.size()) #mean = 0, sd = 0.15 , if sd is of higher value, the clipping would happen , value will be clipped at 1"
      ],
      "metadata": {
        "id": "u4X42-uqHZgV"
      },
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "plt.figure(figsize=(12,6))\n",
        "plt.subplot(1,3,1)\n",
        "plt.imshow(x_train[0])\n",
        "plt.subplot(1,3,2)\n",
        "plt.imshow(salt_n_pepper_noise(x_train[0]))\n",
        "plt.subplot(1,3,3)\n",
        "plt.imshow(gaussian_noise(x_train[0])) #even some noises are added, we dont see any kind of differences, if we DONT do NORMALIZATION"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 262
        },
        "id": "pEGinhSJKvDO",
        "outputId": "6167bdce-9f4c-4937-c09c-0ce7e4db3d54"
      },
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f2cbd9fc6d0>"
            ]
          },
          "metadata": {},
          "execution_count": 11
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 864x432 with 3 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "x_train = normalize(x_train)"
      ],
      "metadata": {
        "id": "oAX7HlNYLLOw"
      },
      "execution_count": 12,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "plt.figure(figsize=(12,6))\n",
        "plt.subplot(1,3,1)\n",
        "plt.imshow(x_train[0])\n",
        "plt.subplot(1,3,2)\n",
        "plt.imshow(salt_n_pepper_noise(x_train[0]))\n",
        "plt.subplot(1,3,3)\n",
        "plt.imshow(gaussian_noise(x_train[0])) #after normalizing"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 262
        },
        "id": "SeGzb-NPMQH-",
        "outputId": "2ed754ac-8789-4e48-87ca-becc56a28852"
      },
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f2cbd8b7e90>"
            ]
          },
          "metadata": {},
          "execution_count": 13
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 864x432 with 3 Axes>"
            ],
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAr8AAADkCAYAAACRz0zzAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3deXxcZ33v8d+j0UijZbRbsizJlu3Yju0sTmyykwVCCFxCCG0CKXuBsKVlK4Wm5bJc6KUty4tSAq8EQgINWxrSOCwhSynZHCeOYydxvNuyLVurte+jmef+YeVi6HyPLVvSjHI+79fLL9v6+sw858x5zjwejb7jvPcGAAAAhEFOpgcAAAAAzBQWvwAAAAgNFr8AAAAIDRa/AAAACA0WvwAAAAgNFr8AAAAIjdyT2dg5d6WZfdPMImb2Pe/9V4L+fZ7L9zErOpm7BF5W+q2703s/Z6bubzJzNjdW5POLK9JmqWjAfaR0FhnW1Yo+1+ks6L/pAW2NyXyd5QaMJTKqdyKVpwcTtO+JIr1/0f6ADY8hEQ8YT1JvFxnV+z9WoscaGQu4zRF9mzmDozIbrY7JLL9H78RoWURmLuC8yOvSOzFamaezluasna9mR+ZsXjz9nI326X32+XpCjxXrcyF3RI8lZ1Q/boli/bhFEgG3Oa4f1GTe1F8/guZz0PnlUsEVsj4nYKw6sqSeJoFzPSfgmPqAVWDQfA463kHHLYjXp4VZwHHJGdPjHOw5mHbOnvDi1zkXMbNvm9lrzKzZzJ52zq313r+otolZkZ3rXn2idwm87Dzk/2PfTN3XZOdsfnGFLb/q42lva2huwEJuUI+hYmvAAqhMX44ShUELTn3h62vU21VtGZdZ0Z4+mQ03xGWWO6ifgVou0M9cdf8dcNCOUcV+8DL9gkJ+t964bLd+Rtx/hX4sivfrY1q+Q99mwfpdMmt633KZLVjbJbN9V6df6JmZOf3wWuOPD8hs79sbZLb9S5/I2vlqZpYXr7BTr0k/Z2sebJb3NbpIr+cPBZy35Tv0+R7fMyCzlleW6u0O6NuMHdbnV3+D/p9u0H/mchJ6juQNBP0H+cQyM7PxgOtZMuCFhd4lervcgEtIUYtejY6W6dss26WPd998fY2IDun9D1oYjwY8TqmAxXbJPj3ZH//Pv007Z0/mbQ/nmNku7/0e7/2Ymf3UzK4+idsDML2Ys8DswXwFpsnJLH7rzOzo/z43T3ztjzjnbnDObXDObUiYftUHwLQ75pw9er6OjwS8lABguk36OZY5Cxyfaf+BN+/9Ld77Nd77NVELeAMegIw7er7mxnh/PpDtmLPA5J3M4vegmR395qj6ia8ByE7MWWD2YL4C0+RkFr9Pm9kS59xC51yemb3VzNZOzbAATAPmLDB7MF+BaXLCbQ/e+3Hn3I1m9ls7UsNym/d+y5SNDMCUmuyczRn3VtiR/qdo5zzaIe9n7/XzZOaS+qeAS7Z2y2y0VjcsRHv1zxJU3H9IZokVC2TW+krdIjBSKSNLxvQldcm39sis6d2LZFa+M6DDyMzm/0oft+H6YpkVbG2V2bwC/RjGn2uR2eDyaplZrW4UCKyiygn4yfh8fT4t+tpmmY0PD8ssEa8PGMzMOZHnWOfNImI69K7Rj2miSB/jxv88LLO+U8tk1nV6icyCKrRKtvXIbKROXweCasCCag2D6hCT0YD2gYBqxvGC4LaHWJee0yWH9Pu2c0f1/ud368aDvDZ9m4dXl8tsLKBGsXKLnkN9jfrBKGzVDRIV6zpl1n5Z0POKjKST6vn13v/azH59MrcBYOYwZ4HZg/kKTA8+4Q0AAAChweIXAAAAocHiFwAAAKHB4hcAAAChweIXAAAAoXFSbQ8AXr5yRsatcGf6mqNkha7QChLtHJLZeGmBzGJ7dAVO8pCu7Nr3idUyKz6o64jm/WeTzJqvbZTZnEd17Vpynu5IiwZ8Km38ty/q0MxGLlgms/46fYmPtZbK7PAKvd1gja4c6m+UkeWvqJLZotv2yWz7xxtktuSzus6s4/oz9WACxDp1hVW2cymz3NH053XuUEpuF9/ZK7PWi3XtX/n2MZmNxfNkFuvWc2+8TF8H+uZHZVbzkK413POO//Gp0P9fYaseS36fPmaDc/Vrh3Oe1dc5M7Pmy/T1s6Fd325OIqCyLaa3G6nT91e+XV98Buv0Y9F5us6qnxmQWSLgvDhwjX6cXEB7XO7w5F/H5ZVfAAAAhAaLXwAAAIQGi18AAACEBotfAAAAhAaLXwAAAIQGi18AAACEBlVnANJK5efa8OL09Vyt5+rKocavPy+zxOolMsvf3S6z8QPNMsutnSuzBd/dKjM/PKzvb9VSmdX9ZJfMbHhERts/v0JmS/72KZnt+/Q5+v7MbMF9XTJrX61rqnZfWyKzyud1xVP3cl0FtuSWgzIb368zv2yRzCIj+v723KTrzBrX6rqlQ5fGZVb3oK79Ci6dyzzvzFKR9Fl+t67h611RJrOy3QmZjVTp60DB4aTM8rt0RVoirm9zeI4+F0YbdZVg5Yt6LHl9OhsrEQfTzMZK9Fh6lhTJzMwsqk9Na71Az8vcYd33VdCp52yiTI+1YG+3zCJVMZlZQCNg93JdreaSARWTj/TJrGO1nrOjAfun8MovAAAAQoPFLwAAAEKDxS8AAABCg8UvAAAAQoPFLwAAAEKDxS8AAABCg6ozAJO28M4WmY2fpmurDrwmX2ZzqupkVjKia5q6Lm2UWd6Arv9J5ut6nEShfl2g8rCu8em5VO97fK++zZErzpJZjm5iMjOzofm6Gql0t97/8qf0Y7j/z/VjEevQYzl4Vb3Mar6tq84GlpXLbNGXN8ts/0dX6dtcUCizZJ6MrO38Uh1u1FE2iIylrPhA+rq9oPqtlG4Xs+igPm9zh/T5lX9Yz9nOVfqxCarsqnxxXGaJYr2cCaozi/bp2rVUVF+vYp36uJT/9BmZmZkdfvtqmZXt0XWJQ9X6xB2cq8dT0qSPW6pUPxaDNbrqrXyHPm65/TprPV9XlqWiOovo0ylwPiu88gsAAIDQYPELAACA0GDxCwAAgNBg8QsAAIDQYPELAACA0GDxCwAAgNA4qaoz51yTmfWbWdLMxr33a6ZiUDg+Llc/fJE5VdNyn9v/plFmyUJdUbNgcbvMCj+sK6dav647TDau+ZnMOpODMjv3rk/K7JRPPCmzl4NJzVlvlpNI/5geel2t3Kx0r67VqXlaVw4V3Pu0zOzUU2RUfEh34ESG9FiG5xbIrOT3u/RYKnQVVuxwQmZ5A/q1hrzegO16gy/T7Wt0Plqlj3fbufoxfMW522T29PqlMov26328ee/vZZZjOrv2S5+SWX6Pl1lRwHnRemFMZlUv6OvYTJvsc+x4LMe6VqQ/ryu26gqt8SJ9Dg3NCagQC6gSHKzXxzhHT0srbA2qHtPnVyKua7k6T9fPI2W79G0W7tfPI6MlupZr4I26utDMzOnT1pov1delkbn6wM2Z3ymzwbv1emDf1fr+/u6Se2V21yFd17Zj11yZlW/SO587rLOga2uiZPJL2ano+b3Me6+POoBsw5wFZg/mKzDFeNsDAAAAQuNkF7/ezB5wzj3jnLthKgYEYFoxZ4HZg/kKTIOTfdvDRd77g865ajN70Dm3zXv/yNH/YGLC3mBmFjP9MXoAZkTgnD16vubHyjI1RgBHTOo5NlqsPyYawB+c1Cu/3vuDE7+3m9k9ZnZOmn9zi/d+jfd+TdT052QDmH7HmrN/NF+jRZkYIoAJk32OzY0xZ4HjccKLX+dckXMu/tKfzewKM3thqgYGYGoxZ4HZg/kKTJ+TedtDjZnd45x76XZ+7L2/f0pGNUtFli+Rmc+PyuzQJfrby8Pn6aqVilKdPXqmrgHLhN8M6VqYf/q3K2W2/vQfy2xvYlhmX2l7jczmPRrQM/PyNvk5Kw5V3S8P6m2crq5re7Wu1xp+z3kyGy/Qt1n7U13LZQldDdRz0Wkyy0k0yqzpOn3+5PToS2pJQKXS333j5zKbn9slMzOzn3afK7PfN+uKuOVVbTJ7+vFTZZYq0fVpiWpdL3b5gx+T2aWnbdf3F/AslYzp86LnFF3hVLxPbzcwV2czbPLz1U38SiNRHFBnVqOz6JA+35P5+lgNVevzfd7tAWv4qB7Lobfp87J8h67C6l+p69Nir+qVWTKiz/XTyptkFs3R25mZ9Yzpc/PP5jwjs0f7dc3gvZtWySyyXI/lG6/Wz7HrBvT1o6W3RGbVjwWcT8O6Hq+/QdfVFbbp83Bgnt5OOeHFr/d+j5mdeaLbA5hZzFlg9mC+AtOHqjMAAACEBotfAAAAhAaLXwAAAIQGi18AAACEBotfAAAAhMbJfsJb6CQvPVtmX7/92zJbGs2bjuFklYTX9S7/+1vvllnuoK4wOf+uG2UWP6hrrPI7dQ1a4Yb1MsMfJOLOml+V/rxddJf+tMaWi/WnTM19sl9mkQPteiyLdUWay9NzKzk/oFrtHF0VmH9Vp76/F+pltvut39Vj8bri55T7Piizd5//mMzMzO7deJbOXvMtmb1p7UdlVvdkQB1RvX7amPOsrpva+ya9XesNdTJLXKHrtIbn6nEO1ertlnxP17wld+yW2WyQiqTf74F5+vhXP9Yhs7ZL5sisdI+uEIs36XNhfOVCmXWu0teWmqv3y2wgoa8D76zVVXpb++fKbG9PpcxubXhcZjce1PWDZmY5Tj/n/cuuK2Q2mghYsiX0a5mFLXoufOrud8gsWajnV3GTrhfLG9BrgbG4HuecZ0dk1rtIf0ha7YP6ueM58XVe+QUAAEBosPgFAABAaLD4BQAAQGiw+AUAAEBosPgFAABAaLD4BQAAQGhQdTZJ+dsPyeyZkQaZLY3qap2Z9smW8wLzPQNVMrt98X/IrDel61tq/vWJYw9sCumR4HjlDplVb0xfdZMs1LVCtd/fJLPhS1bKrPvCxTKr+62uYhpbrKuKolt1NdInztT1R6fmt8js4/3XySyo4uj8kl0ya1ysrw93vvgKmR3L8mhUZr5Y1xEVrdWPYdGZy2Q2Uq1rqvI79Wstg4tKZFb97KjMEnFdt9Ryvs46LqqWWe97dGaf0de/bJAzblbUlv5xTRTq43/gar3PNU/p+imX1FfaZIFeXgxX6fMyUaRruSpjup7wgqo9Mmsb0+dXrtN1XlfPV0VZZm9vulRme/sqZGZmNhJQWXZ5/Q6Z9Y/HZLbut7qGNTKmH6fynfo60L1UjzOiTwuL9usa0v4GXVnmc3SWO6z3YaShVA9mW/ov88ovAAAAQoPFLwAAAEKDxS8AAABCg8UvAAAAQoPFLwAAAEKDxS8AAABCg6qzSRpvaZXZt/7pWpl9+Upd0RJ5rlhmmz/8reMb2J/4UucZMtt1ua4jMjNL9uiap784/8Mya/prfZsLbXPgfSL7RPqHLf7Q1vRhQ63esF5nsQefldm8g6fIrOeMSpl1rdTVSOPXLJXZv9yvM69bssxV6OqtysYBmRXl6O3eXr9eZnd/9GI9GDMbatS1c6u3/pXMigP2MXneaTLL3bxbZgXj9TIrKdN1RKmIfgxHKnUtVlLvuhW26tuMDup6q0rdbmV6z7NDZCRp8Z29abO+U8sCttPHaqBeH+SK9e0y616j69OCqrdqNugOrT2tumZv05vqZOae0VVnZ131osx+0XSmzIrzx2QW/2yRzMzM3AL9HHx/3QUy6z9LX0OqhvQxTQXM9dHSgNdAAzpDS/brOrOuFbqyLMhoqT4P4wf1nE3EJ7+U5ZVfAAAAhAaLXwAAAIQGi18AAACEBotfAAAAhAaLXwAAAIQGi18AAACEBlVnU6jiB+tkNuc+XdWUPNwls5Wn/aXMtlx8m8zW3nKJzKp7npDZsbh1urJsod59zEKJygI7dL2ovAqowKnYriuAIjW6ciivqUNmOUldk7X4B7qa7/CFunatbEu/zCIdPTLb9cH5Mju0VNdJ/XTbapnlbNV1hwu9Pi5mZkXbdB57WB+b1Nmnyix3xwG93VK9/znDCZmZbjGyruW6i6lsl644CnoMy5NJme1+a7nMGh7W52+2S8Yi1r80/VwZnKtf66p5Slf0+YjeruscXWcW36cry9rO1VVfba/Q9zdnk77wxNfGZZbM19s9vllXHv7DpWtl9q2b3yyz8cX6nDUzSxToyVByQJ+31Zv07Y6W6e1Sufr+ehfquVd/v16b9K7U17o5m4dl1r0kJrNYj96/6IDOUtGAi4twzFd+nXO3OefanXMvHPW1Cufcg865nRO/6ysJgBnFnAVmD+YrMPOO520Pt5vZlX/ytc+Y2cPe+yVm9vDE3wFkh9uNOQvMFrcb8xWYUcdc/HrvHzGzP33t+2ozu2Piz3eY2ZumeFwAThBzFpg9mK/AzDvRH3ir8d6/9GayVjOrUf/QOXeDc26Dc25DwvRH8wGYVsc1Z4+er+PD+iO5AUyrE3uOHdXv3QXwByfd9uC99xbw4y/e+1u892u892uidmKf9wxg6gTN2aPna25B8OfTA5h+k3qOzdc/OAngD0508dvmnKs1M5v4vX3qhgRgGjBngdmD+QpMoxOtOltrZu8ys69M/H7vlI3oZSrZefiEtkv05Z3Qdivf9qLMOr6jq03MzCylK1Mwa016zkaGvVVuSf9WpdZz9XdxYnt1PU7HK+V3by2nsUFmBYfHZbb9Rl1ntuSHvTLb+U5dnxZr19mif9wks5aL9HYL5+hrwECp/nb17lx9zMzMUlGdRUb1sSndpbfrvnqZzE75WsCGFXr/y17Q9XFdK3SZQc9i/RrNwDx9fwndqmfzHtOVbNG+rKg6O7HnWG/mRCNU7rCu+xqt0PM5v1NXlkXG9G2m8vXzTPVGfZuJYr0sGa7UtzlQp+uuxsp1TZYb0+fXtmE9f8rfeFBmvXfPk5mZWc8KPZ7GtfpaN1yl1wN5/fp5O7dfn+99C/R3+FLF+v7G4vp4p6L6fBoMeJxKmvS+98/XY4kOBnRvCsdTdfYTM1tnZsucc83OuffakQn5GufcTjO7fOLvALIAcxaYPZivwMw75iu/3vvrRfTqKR4LgCnAnAVmD+YrMPP4eGMAAACEBotfAAAAhAaLXwAAAIQGi18AAACExolWnWGGLP/0Dpm953T98xA/WPCwzC659iOB9xn/2ZPHHhhe9lzSy9onH9FVNm5M1+rk6Mg6V+u6mrxu3eeVpxu0rOViXaE1Z4OuG8rr15U7OSVxmQ3/H11x1LFKH7MPv1c3Wd31z1fKzMws78mtMmt9zyqZlW3rl1nFJv1AuZjej2TTAZnZaafIaNH/fU5mTX9zpsyCKrOaX6XHOVCnn/qKtgScUFnO55glCtNXSZXu1hVueYeHZDYyV1dhRRJ6zg7V6DlbvF8/bqlcXYVV1BJQUTeoa9CCbjOZr7NNp9XL7MI5e2R2T3mdzMzMKp7XrzsOV+vjFh3S16ygfQyqnat9QtcsDtUWyKx6na60HFqgewYrHzsks0Ov18e7sFPve3zP5D/ZkFd+AQAAEBosfgEAABAaLH4BAAAQGix+AQAAEBosfgEAABAaLH4BAAAQGlSdZblkT6/MDn9oucz2rx2W2We+9MPA+/y7666RmX+2VGYNX16nb9TrShxkJx9xlijJS5s1fHm93K7zbefIrPL+XTq7d1RmXW9cobd7aK/MkvVzZOYSSZnltHfLzJfqqrOexemPl5lZ1Wa9f79sP0Nme68Jvkzf+K+tMlv7aV0N1b9IV1h1nqlrk+Y9pve/d8F8mQXVYkVW6v1f+M0tMjtww0qZFbTLyEYqdLb3nQ06/KKOsoHPMRsvSP/YtZ2jq98KW/R5m9+vK6ZG4/r1s6I2XRc40BCTmQt4qmhfrWvAFt6r666SMT2Hhubqfe//ga4si396m8xiF3TKzMysa5+uYKzYrI9p7LB+LAqadXVh70p9f9EBXYPmA14e3fkuPYmW3KKvSX1n18qssENfkxNFejA5g/raKreZ9BYAAADALMXiFwAAAKHB4hcAAAChweIXAAAAocHiFwAAAKHB4hcAAAChQdXZLJbavFVmb/3Cp2R25+e+Gni7m84LqEI7T0cri26U2ZJbW2Q2vqcpcDzIjETc2cFL0tcAzY2dLbereqRZZk3vWyKzBXe3yazimcMy2/GJRTIrPKgru+Y8NyKz/ICqs/6VlTILqgZqvlxXKsV/0iizN97wtL5RM3uo41SZ9d4QUH/UWygzP6LrjwqfapJZtK9eZsl8fXAKdgdUQxXpSra5TwzJ7NAr9f5VvqhruPJ6dbZdJtnB55glitKf81Wbde1dIq4f765lOpv3mK7U7GvUdWYp3Vhm0UHdddb4y0GZdZxVLLOus3WFVtnzAft3926Z3bbscpktv3iPzMzMfn7VbTL71Bm6anTrQ/r6WRMpCbxPpWjzQZkNr5wns/qH9TEdWlIls6B6vKqn9HWg43x9m+0X6sxeTP9lXvkFAABAaLD4BQAAQGiw+AUAAEBosPgFAABAaLD4BQAAQGiw+AUAAEBosPgFAABAaByz59c5d5uZvcHM2r33p0187fNm9n4z65j4Zzd57389XYPE5FXctk5mN27/SOC2JV/RPa0/WfRbmW1557/J7NSG98ls2Rf0/8GSO4P7EvE/TdWczTs0aAs+92TabPDN58jtUp1dMitq0T2wQdov0D2OS7+rO6Sbr9Y9lS3n6R7SovkLZVayf1RmY0X5MivbJiMbqdR9xFs/skJvaGbdn9Vdq7E83e26bHGTzHpGC/R4vqCPTeO9KZkVPqevK90XN8psoF5fHyL6obDxYt0Xe3ilfupzqYCnxd/p6ERN5XNsTtIsdljsd8BLXUPVAd2rL+jeYx/R522Qsh36nM0Z0/2xPct0l2/BYX3u1T2oxzk0R58nHVfqDvHaJ/Vx2d23WGZmZq9d9Vcyu/fCm2X26Ve9WWYvLtLXuvgmXaycd3qdzMYCOnnje3XncusFunM41qUfp4Gl5TJL6kur1TzRo0PheF75vd3Mrkzz9W9471dN/GLhC2SP2405C8wWtxvzFZhRx1z8eu8fMTP9Ug6ArMKcBWYP5isw807mPb83Oueec87d5pyTr1U7525wzm1wzm1IWMD3qABMt2POWeYrkDUm/Rw7PqK/FQ3gD0508fsdM1tsZqvMrMXMvqb+off+Fu/9Gu/9mqgFvGkDwHQ6rjnLfAWywgk9x+bGimZqfMCsdkKLX+99m/c+6b1PmdmtZqZ/+gVAxjFngdmD+QpMrxNa/Drnao/66zVm9sLUDAfAdGDOArMH8xWYXsdTdfYTM7vUzKqcc81m9jkzu9Q5t8rMvJk1mdkHpnGMmGLu8U2B+dCfV8vsFW/RFS3rP/1NmW277Hsye1vjFTLrvUhGEKZszjpnLi8vbRTf3S83673qdJnFD4zp++vuk1HFtrjM9r5NV/yU7dS1OqXb9P11rCmVWTI/IrP+Rl2pVHxAVyoFtWvtekuhDs2s/ru6sq3zdH3DG0/XtVFPXKTrll75yKdkVvicriZM1VTIrGSPfq9q3oDevwOv1o/F4rt0ndZ4XFc/dZyZ/pyfLlP5HOsjZqPl6c/BeLOuEKtZNySzVL4+h6KtvTLLHdDn10iNrtIzrx+b0j36MY0e1D8zePhCfY2o3qCvZf2L9D4UPacrFlO5+v7MzAo69Dl93ZZPyuwN1z4hs9Vn7ZfZ+vmNejCPlsko1qKvn4ON+ppc1Kq3K/55+vpMM7PUJWfpsbTrn0HpP0VXq5lY7hxz8eu9vz7Nl79/rO0AZAZzFpg9mK/AzOMT3gAAABAaLH4BAAAQGix+AQAAEBosfgEAABAaLH4BAAAQGsdse0D4JNvaZVbzrzob+dtxmRU6XR90a+MvZfaGaz6mb/Oe9TLDyRutK7TdHz07bVZ4SFd65fXpSq+yTXv1HdZUyWjv1boaqfFXuv5ouEqfd6lNL8qsemixzFzfgMxijYtklgxo0Gq4+Xl9f4UBtVBmtucjeqwL/3GjzJp+tERm8Rw92POu1GN9/NSFMqv7gb7Ngg26Iq35umUycyl9rrVcqD/tLEdfqmxonq5pynaRYW8VL6avhMrfdkhuN7KiTmadZ+hPeizbqbPCJl0l2F+vlx6lu3Ud4kC9rggbW673IdajH9PhWl0lGN+pa9CSNboirL8heGk1Z6Ou9hst1edt+6iuF7ux5mGZdSf0Pv76+gaZ1TwlI4sdTsisaJ++Ro69erXMRsv0cRsp16/VFrXpGj+FV34BAAAQGix+AQAAEBosfgEAABAaLH4BAAAQGix+AQAAEBosfgEAABAaVJ2FUOqiVYH57mt1ncxpq5pkFlRnFuRbXWfp27x3wwndJk5eTsIs1p6+0qz2MV1j5Lbv01lZqcxaLqmUWSqqK62irbqOaLBW16dFluuqr9HaEp2V69vMCWjc6V6p96GqX++DrdT1YWZmKy7dJbPCy/Vc/k3jj2R20XNvkVlLp34MS9bpWraxEn1wokt13VK0X9fq1T6p65aS+fq1ncGaiMyqNur7a5JJdvARs/Hi9PvWcrU+j6JD+jbrftsps+EGPU9ShVGZ1Tyib7PnTH0dKH+8WWZDK+bKbKRSL3Uio3peBr08OFSv68Oi/QG3aWa7rtfzsrxRH5v5BV0yW9unn0d/d0Bf66qelZFF+/Wc7V2on+8rhnWXYLQ3fRWfmdlAnb7N6nv1da77cl33qPDKLwAAAEKDxS8AAABCg8UvAAAAQoPFLwAAAEKDxS8AAABCg8UvAAAAQoOqs1nMrTlNZjv+WleG3HrhHYG3e3Fs7ITHpIx6XUn0ZFdAlVOqZcrHguOTkzArakmlzfoXFsnt2q/V52XpDn1/c+/cIrPaUl2p1HN+vcyKmkdk1r1KVyoVN+s6HhfQYpQ7rMNUXvpjaWa260e6pijnoK5FMjPbvOhXMnvfvtfKbOW6t8nsd+fcIrPz1n5CZnkBFU+lD26TmeXnyyhZoGuMDl6s67QWf03fX5HXj8WBG1bKLOs5Z+Oi4q20SddPDc7VS4G+FeUyC6oJ612qrxEVTw7ILL5nUGZjjXNklhAVb2Zm4zFdX1d0SB+XtvN1rd/IJbqe8LLGnTIzM+sc1cdmaFw/dz/dtUBmF32vIPkAABIxSURBVFTukdnwbn39rOjVc8FH9HEr3avXCUO1+prV36Afp9rH9THtuOoUmY0X6nEqvPILAACA0GDxCwAAgNBg8QsAAIDQYPELAACA0GDxCwAAgNBg8QsAAIDQOGbVmXOuwcx+aGY1ZubN7Bbv/TedcxVm9jMzazSzJjO7znvfPX1DffnKXajrS3a/Z57MPv+Wn8rsz4o7T2pMJ+KmtjUy+/03z5NZ+R3rpmM4oTSV8zW3d8QqfrU1bebra+V2Q9UVMssb1LU6A5edKrOgyp34Hl2b1HKRrvgZrNM1TbEuXcczGtevGfiAlxNWn66riD5e94DM7jx8gb5RM/tsm55b617Q9UAfvvC/ZPb6L/6NzHKW6eM2UB9QjbRSVxoOzdVVZwtvelJmI294hcz237BcZvndeh8iuh1vWkzlnM0ZTVrJDlEXFTCHUtFCmfXX67mQDKgQq32kV2bjc/S87Fita8CKWpMyGyvSky86qB/vQ6/UtVyrrkx//TMzGxjX52xejq5PMzM7o+SgzKqjfTJ7pHupzO75t8v0bfbq/S/a2SWzrjVVMsvr1/WlyXx9XtSvPSSz0fn6ucMCKiZr1uuKNOV4XvkdN7NPeu9XmNl5ZvYR59wKM/uMmT3svV9iZg9P/B1AZjFfgdmFOQvMsGMufr33Ld77jRN/7jezrWZWZ2ZXm9lLn5Zwh5m9aboGCeD4MF+B2YU5C8y8SX3Cm3Ou0czOMrP1ZlbjvX/p47da7ci3bNJtc4OZ3WBmFjP9rRUAU+uk52uO/vYjgKl30nM2T38iGYA/OO4feHPOFZvZ3Wb2Me/9H70pxXvvTbwjw3t/i/d+jfd+TdT0e2QATJ2pmK95rmAGRgrAbIqeY3N5gQk4Hse1+HXORe3IpLzTe/+LiS+3OedqJ/JaM2ufniECmAzmKzC7MGeBmXXMxa9zzpnZ981sq/f+60dFa83sXRN/fpeZ3Tv1wwMwGcxXYHZhzgIz73je83uhmb3DzJ53zm2a+NpNZvYVM/u5c+69ZrbPzK6bniHOHrmN82XWu1pXQ73li/fL7INlv5DZdPlkS0B10s26zqzi9qdkVp6izmyGTNl89cmkJXvS1xUd/vMVcrviFl1HFFQF1jdfX47q7muWWbJK1yYFNQ4tu1lX7nSfq+frlz77PZntHJsrs3/fd47M3n7/B2XmUro2yMxs3ikdMpv/S73tz566QmbjhXq7RfcMyywyOCazfW8ok1nJXl2BF6kol1nRYztl1r1UV51V//tmmbW/40yZTZOpe451znx++mqy8aKo3iyle6SC5lBRk57rHav1vAyqHivfNiqzvgV5MnMBVVj572uR2f+q3C+z53t01eippW0yK8zR88DMbEO3rjfdsn6RzOr/Sz8Y5aP6uI1U6Mf+wBurZTZns96P3dfpx2Lp7YMy2/9mfUwLOvSDGB3WWaJEj0U55uLXe/+Ymakr4asnfY8Apg3zFZhdmLPAzOMT3gAAABAaLH4BAAAQGix+AQAAEBosfgEAABAaLH4BAAAQGpP6eOOwyK3VdUVdt+mPfP3Qwt/L7Pq4rkWZDjcevEhmG7+zKnDbqv94QWYV/VSWhUVibpE1/+UFabPYYV07E/+vbfo2z9Q1Pt1L9CfK7X1Hvcyqn0nIrHeNrv/pXabneVF9n8w2DOl9qImmr4YzM/v44odk9tmNfyGzOc/qGjAzs9bhtJ96e2TbeECFlW6psnkP6fq0tourZFbUpiuHIiP6/uL7dOgK9HnRfK2ujKq/VV/Huq45Q2bVT/bILNsl83Osb1H6T3krPhBQhVWpq7CKAqoLk/m6Eq9yi67E6zxdP6aHT9fn0LWvfVxm1Xl6zt514GyZJXz6ajiz4Dqz8ZTe7oGbL5SZmZkLmNINB3S9WN5hPU+G6/TapLhpQGbjsbjMknn69dH5v9Q7kYjrx7DhPv1ZLcONug6xv0Gfo4mCyVed8covAAAAQoPFLwAAAEKDxS8AAABCg8UvAAAAQoPFLwAAAEKDxS8AAABC42VddTb22jU6+3iXzG465dcyu6Jg8KTGNFltSV0Xc/HaT8rs1H/QdVMVPcF1ZcHFSgiLVNRspDr92VC6R2+3/XPLZVa2VVcjJUp1LdfbXqdrBA9frSt+Fo7HZFYX05VWV5Q8L7N3PfR+mRVVDckskqNn1tx1uk4qoInJzMyKD+hjWvJjPdcjy5fIbHiBrhzyAS+Z5A7pfazcqivpRqt0VVHebn1s6u5rkdnuT62UWePnnpLZ0OtXy8w26Sgb+ByzRGH682G0QldFxffq5zUf0efXjvfmy2zBAj2/Pjb/lzKbH9XPzV/Yc5XMrqrVc3ZxaafMavN0PeF3H7tMZgvu09erim59HTAzG6zX16VUQL1YqlA/htGBcZkNNehrZPlm/TglS/TjGxnQ1Xnt55fLrHK0WGaFO/XjNFquqykjo/qxUHjlFwAAAKHB4hcAAAChweIXAAAAocHiFwAAAKHB4hcAAAChweIXAAAAofGyrjprepNe2+84/a4pv79v9yyW2Td/f4XMXFJXyZz6pb0yW9K2Xma6HAg4PtF+b3X/nb66qninrsdpX1Mps6Fafa4XNeux3PH7V8rMVYzJrLqyT2avXfSCzN79mw/o+wuYXPO+pquIzOs6nsE6fVzK1gUcGDNzqTqZ5cR0pZIbGpFZzyl6P0ZL9VgK7jkgs1T/gN5wcYOMei9YILPooK5Wy+vVx3T0dWfLrPB3W2SW7SJj3uIH0lfKHV6p6+Si/TqLDOsKrZxB3cP3iYUPyOyZoYUye7L/FJl5rx/Tbz+gn2Nzh/R2z7WfJrOGfXrfB+fq5dNYSXA/oQvoE4116HkZsPs2PE/XkuUEXLOGF8QD7k/fYcerdH1ayV69g0FVbh2vrJWZBex7QYeuUVR45RcAAAChweIXAAAAocHiFwAAAKHB4hcAAAChweIXAAAAocHiFwAAAKFxzKoz51yDmf3QzGrMzJvZLd77bzrnPm9m7zezjol/epP3/tfTNdATsfRDT8nsDR9aPYMjMVtqeixBqCzDZEzlfM0Z95bfk75CJrV7n9yu6GCVzOY91CUzv32PzNx79XytDGimirboyp0vXn+9zOIBrVyjZTo7eHGhzOb/RtfDRRK6pmj3+3XVl5nZvMd11dvQ5WfIrPly/drH0u/pxymns1tm2z6jK6zyugJeawmIUrm6Iq74gN6wfKeuqSpcr881a6zXmW7HO2FTOWd9xNlYafqn9eJDei70Neqqs/Kt+jiW7NTH/6ufervMOs7QS4/ooIyseqOuAVsQ0XVX+Z3DMvNRXUs2WqHnZdUGXaOYqNTXATOzjjN0BaF3ukKs/JEmmRUUzZNZz2K9H5Ex/RgWdOjHft5j+rHob9D3N1irz7WgKsyg+rShmoCKSeF4en7HzeyT3vuNzrm4mT3jnHtwIvuG9/6rk75XANOF+QrMLsxZYIYdc/HrvW8xs5aJP/c757aamW5VB5AxzFdgdmHOAjNvUu/5dc41mtlZZvbSR4vd6Jx7zjl3m3OuXGxzg3Nug3NuQ8JGT2qwAI7fyc7XsUTA9x8BTLmTfo4dZc4Cx+O4F7/OuWIzu9vMPua97zOz75jZYjNbZUf+1/q1dNt572/x3q/x3q+Jmn4fCICpMxXzNS+q33sGYGpNyXNsPnMWOB7Htfh1zkXtyKS803v/CzMz732b9z7pvU+Z2a1mds70DRPA8WK+ArMLcxaYWcdc/DrnnJl938y2eu+/ftTXa4/6Z9fYtPwcLIDJYL4CswtzFph5x9P2cKGZvcPMnnfObZr42k1mdr1zbpUdqWZpMrMPTMsIAUzGlM1Xl0ha3oH0tVbDl5wut4s364I+H9OXnJyyUpnV/vaQzHZ8QFf85CTmyqzxs0/IrPft58nMO/2aQSqgcccdaJVZzrzF+v5ydNWXmVnHGbo6qOYZXUfkxvV2TW+ukNnCH+uf3SjZpY9N7QNtMmu+qkZmebohzqrueFpmOcX6LQDD5y6RWWBt0vQsP6duzia95fekr6dK5eoaqaFqvc9dp+narpJ9ugrLpfR5W9gWcE4HRL0L9VsnnW7CstGKgP1brs/Z0t36RsdK9PUqry+4pLTu/naZ9Z6pqyKT9XNkFnS84836cepdqK/JVQ/sl9nAK3QFY+6IPm6F7bqacaSiQGaxw7rKLndI75/c5lj/wHv/mJmlmzVZ1ekLgPkKzDbMWWDm8QlvAAAACA0WvwAAAAgNFr8AAAAIDRa/AAAACA0WvwAAAAiN46k6AxBCidKoHXxD+hoxF9AsM+8+XY/Te26dvr/lp8gsqMYnv0dXOM2/WXdTuQUNMhuu0q8LlO/QO983X19SD77jVJlVbxiSWcWLEZmZmVU8qo9322vny2zuOl1HFH/geZnt/PvTZFb5vL7Nne+tltnS7zTLLFWsq7aGX3uWzAoO9MssMqrHOVKhz6dsl4w561qWvsIuv1fPobKdukYqv1tX241UxWQW7dfzpKhNH+NYu76/8UI9v7pW6Bq0iuf7ZGa+WEZjcX0dqHz0oL7JQn1czMxG63RNWtC52bdI1/flDeh6tUSRvoYUter7C6ozGy/QxyZ3SN9mf4N+nIpa9T50nK23K24OqCd8PP2XeeUXAAAAocHiFwAAAKHB4hcAAAChweIXAAAAocHiFwAAAKHB4hcAAACh4bzX9SdTfmfOdZjZvom/VplZ54zd+bFl03gYS3ovx7Es8N7PmYLbmXJ/Ml/NXp7HfyowlvSyaSxmUzOerJ2vZln9HMtYtGwaz8txLGnn7Iwufv/ojp3b4L1fk5E7TyObxsNY0mMsmZVN+8xY0mMsWraNZ7pl0/4yFi2bxhOmsfC2BwAAAIQGi18AAACERiYXv7dk8L7TyabxMJb0GEtmZdM+M5b0GIuWbeOZbtm0v4xFy6bxhGYsGXvPLwAAADDTeNsDAAAAQoPFLwAAAEIjI4tf59yVzrntzrldzrnPZGIMR42lyTn3vHNuk3NuQwbu/zbnXLtz7oWjvlbhnHvQObdz4vfyDI7l8865gxPHZ5Nz7vUzNJYG59zvnHMvOue2OOc+OvH1GT82AWPJyLGZadk0XyfGk7E5y3yVY2G+ZpFsmrPM18CxMF8zNF9n/D2/zrmIme0ws9eYWbOZPW1m13vvX5zRgfxhPE1mtsZ7n5FiZ+fcxWY2YGY/9N6fNvG1fzazLu/9VyYuXOXe+09naCyfN7MB7/1Xp/v+/2QstWZW673f6JyLm9kzZvYmM3u3zfCxCRjLdZaBYzOTsm2+ToypyTI0Z5mvcizM1yyRbXOW+Ro4ls8b8zUj8zUTr/yeY2a7vPd7vPdjZvZTM7s6A+PICt77R8ys60++fLWZ3THx5zvsyImQqbFkhPe+xXu/ceLP/Wa21czqLAPHJmAsYcB8PQrzNT3ma1Zhzk5gvqbHfM3M4rfOzA4c9fdmy+yFyZvZA865Z5xzN2RwHEer8d63TPy51cxqMjkYM7vROffcxLdtZuRbREdzzjWa2Vlmtt4yfGz+ZCxmGT42MyDb5qtZ9s1Z5utRmK8Zl21zlvkajPmafixm03hs+IE3s4u892eb2evM7CMT35rIGv7I+1Iy2Uf3HTNbbGarzKzFzL42k3funCs2s7vN7GPe+76js5k+NmnGktFjE2JZO2eZr8xX/A/MV435qscyrccmE4vfg2bWcNTf6ye+lhHe+4MTv7eb2T125FtGmdY28T6Yl94P056pgXjv27z3Se99ysxutRk8Ps65qB2ZDHd6738x8eWMHJt0Y8nksZlBWTVfzbJyzjJfjfmaRbJqzjJfNearHst0H5tMLH6fNrMlzrmFzrk8M3urma3NwDjMOVc08QZrc84VmdkVZvZC8FYzYq2ZvWviz+8ys3szNZCXJsKEa2yGjo9zzpnZ981sq/f+60dFM35s1FgydWxmWNbMV7OsnbPMV+ZrNsmaOct8DcZ8zeB89d7P+C8ze70d+WnU3Wb295kYw8Q4FpnZ5olfWzIxFjP7iR15ST9hR96b9V4zqzSzh81sp5k9ZGYVGRzLj8zseTN7zo5MjNoZGstFduRbLs+Z2aaJX6/PxLEJGEtGjk0GztGsmK8TY8nonGW+yrEwX7PoV7bMWebrMcfCfM3QfOXjjQEAABAa/MAbAAAAQoPFLwAAAEKDxS8AAABCg8UvAAAAQoPFLwAAAEKDxS8AAABCg8UvAAAAQuP/AeTQYdKu1HXzAAAAAElFTkSuQmCC\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "x_train_sp_noise = salt_n_pepper_noise(x_train) #x_train is tensor here, big brother of array \n",
        "#x_train_gauss_noise = gaussian_noise(x_train)"
      ],
      "metadata": {
        "id": "A9hPwDg3MXix"
      },
      "execution_count": 14,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#input = x_train_sp_noise (with noise), output = x_train (de-noising) which is required to be supplied\n",
        "batch_size = 512\n",
        "train_loader = DataLoader(TensorDataset(x_train_sp_noise.reshape(-1,1,28,28).float(),\n",
        "                                        x_train.reshape(-1,1,28,28).float()),\n",
        "                          batch_size=512,shuffle=True)"
      ],
      "metadata": {
        "id": "imfCwCRuRf5D"
      },
      "execution_count": 15,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class AE_CNN(nn.Module):\n",
        "  def __init__(self):\n",
        "    super(AE_CNN,self).__init__()\n",
        "\n",
        "    self.enc_cv1 = nn.Conv2d(in_channels=1,out_channels=32,kernel_size=3,padding=1)\n",
        "    self.pool = nn.MaxPool2d(kernel_size=2, stride=2)\n",
        "    self.enc_cv2 = nn.Conv2d(in_channels=32,out_channels=64,kernel_size=3,padding=1)\n",
        "\n",
        "    self.dec_cv2 = nn.ConvTranspose2d(in_channels=64,out_channels=32,stride =2,kernel_size=2)\n",
        "    self.dec_cv3 = nn.ConvTranspose2d(in_channels=32,out_channels=1,stride =2,kernel_size=2)\n",
        "\n",
        "  def encoder(self,x):\n",
        "    x = F.relu(self.enc_cv1(x)) #size = 28x28\n",
        "    x = self.pool(x) #size = 14x14\n",
        "    x = F.relu(self.enc_cv2(x)) #size = 14x14\n",
        "    x = self.pool(x) #size = 7x7 #at the constraint or bottleneck part, image size is 7x7\n",
        "\n",
        "    return x\n",
        "\n",
        "  def decoder(self,x):\n",
        "    x = F.relu(self.dec_cv2(x)) #size = 14x14\n",
        "    x = self.dec_cv3(x) #size = 28x28 \n",
        "\n",
        "    return x \n",
        "\n",
        "  def forward(self,x):\n",
        "    x = self.encoder(x)\n",
        "    x = self.decoder(x)\n",
        "\n",
        "    return x"
      ],
      "metadata": {
        "id": "DgA2QmWNSZp7"
      },
      "execution_count": 16,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model = AE_CNN()"
      ],
      "metadata": {
        "id": "mks5sPYTjbLq"
      },
      "execution_count": 17,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "list(model.parameters())"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "zr3bRwU-kjAu",
        "outputId": "c202884a-e4db-45ef-d1c9-c6e14ecb68f0"
      },
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[Parameter containing:\n",
              " tensor([[[[ 0.2621, -0.0634, -0.0238],\n",
              "           [-0.1955,  0.1217,  0.2712],\n",
              "           [-0.1256, -0.1023,  0.1307]]],\n",
              " \n",
              " \n",
              "         [[[ 0.2566,  0.0421,  0.2323],\n",
              "           [-0.3316,  0.2411, -0.1513],\n",
              "           [ 0.2022,  0.1113,  0.2608]]],\n",
              " \n",
              " \n",
              "         [[[ 0.3140, -0.2720,  0.2620],\n",
              "           [ 0.2334, -0.0635, -0.2060],\n",
              "           [-0.1623, -0.2137,  0.1466]]],\n",
              " \n",
              " \n",
              "         [[[ 0.2641,  0.1388, -0.0605],\n",
              "           [-0.1469, -0.0220, -0.1952],\n",
              "           [-0.0544, -0.1853, -0.2046]]],\n",
              " \n",
              " \n",
              "         [[[-0.2197, -0.2884,  0.2587],\n",
              "           [ 0.1775, -0.1847,  0.1736],\n",
              "           [ 0.1336, -0.1533, -0.2579]]],\n",
              " \n",
              " \n",
              "         [[[ 0.1717,  0.2783,  0.3027],\n",
              "           [ 0.3204,  0.1715, -0.0044],\n",
              "           [ 0.0944,  0.1540,  0.1370]]],\n",
              " \n",
              " \n",
              "         [[[-0.2954, -0.0170, -0.0553],\n",
              "           [-0.1737, -0.1672,  0.1002],\n",
              "           [-0.0764,  0.2882, -0.2354]]],\n",
              " \n",
              " \n",
              "         [[[-0.2993, -0.1521,  0.0414],\n",
              "           [-0.2391, -0.0083, -0.1383],\n",
              "           [ 0.3042,  0.0190,  0.0515]]],\n",
              " \n",
              " \n",
              "         [[[-0.0456,  0.0340,  0.2350],\n",
              "           [ 0.2331, -0.0882,  0.2793],\n",
              "           [-0.1926,  0.0252, -0.2824]]],\n",
              " \n",
              " \n",
              "         [[[-0.2249, -0.0547, -0.0157],\n",
              "           [ 0.0774,  0.0661, -0.1197],\n",
              "           [-0.2773, -0.2663,  0.2043]]],\n",
              " \n",
              " \n",
              "         [[[ 0.1124,  0.1297, -0.0599],\n",
              "           [-0.2128,  0.1104,  0.2934],\n",
              "           [-0.0264,  0.0153,  0.1559]]],\n",
              " \n",
              " \n",
              "         [[[ 0.0099,  0.0229, -0.0886],\n",
              "           [ 0.2349, -0.2230,  0.0780],\n",
              "           [ 0.1454,  0.1791,  0.0566]]],\n",
              " \n",
              " \n",
              "         [[[-0.2949, -0.3110,  0.0803],\n",
              "           [-0.1561,  0.2162, -0.2649],\n",
              "           [ 0.2766, -0.3219, -0.0007]]],\n",
              " \n",
              " \n",
              "         [[[-0.1091,  0.3242, -0.2168],\n",
              "           [-0.2386, -0.1172,  0.2838],\n",
              "           [-0.0124,  0.1192,  0.1416]]],\n",
              " \n",
              " \n",
              "         [[[ 0.1530, -0.1009,  0.2531],\n",
              "           [-0.2719,  0.2028,  0.1028],\n",
              "           [-0.0348, -0.0826,  0.1018]]],\n",
              " \n",
              " \n",
              "         [[[-0.1894, -0.2887,  0.1205],\n",
              "           [-0.1859,  0.1102,  0.1319],\n",
              "           [-0.2801,  0.2534, -0.1722]]],\n",
              " \n",
              " \n",
              "         [[[-0.2480,  0.2955, -0.2957],\n",
              "           [ 0.2687,  0.1318, -0.0847],\n",
              "           [-0.2812,  0.1648,  0.1244]]],\n",
              " \n",
              " \n",
              "         [[[-0.0590,  0.2806,  0.0974],\n",
              "           [ 0.0232,  0.0537, -0.3141],\n",
              "           [ 0.2309, -0.2436,  0.0020]]],\n",
              " \n",
              " \n",
              "         [[[-0.2084, -0.1289,  0.0120],\n",
              "           [-0.1369, -0.2658,  0.1207],\n",
              "           [ 0.2347,  0.2767,  0.1865]]],\n",
              " \n",
              " \n",
              "         [[[-0.1363, -0.0521,  0.3173],\n",
              "           [-0.3111, -0.0360,  0.0655],\n",
              "           [-0.1229, -0.0081,  0.1810]]],\n",
              " \n",
              " \n",
              "         [[[-0.2069, -0.0996,  0.1715],\n",
              "           [-0.0706, -0.2485, -0.1669],\n",
              "           [ 0.1335,  0.2788,  0.1248]]],\n",
              " \n",
              " \n",
              "         [[[ 0.1164,  0.0983, -0.1756],\n",
              "           [-0.1761,  0.0792, -0.3267],\n",
              "           [-0.1574, -0.3005, -0.1939]]],\n",
              " \n",
              " \n",
              "         [[[-0.2365, -0.2253, -0.0436],\n",
              "           [-0.1116, -0.1134,  0.1191],\n",
              "           [-0.0227,  0.0721,  0.2789]]],\n",
              " \n",
              " \n",
              "         [[[ 0.1034, -0.0405,  0.0347],\n",
              "           [ 0.0744,  0.2804,  0.2593],\n",
              "           [ 0.0878, -0.1348,  0.2822]]],\n",
              " \n",
              " \n",
              "         [[[ 0.0443,  0.0150, -0.2343],\n",
              "           [ 0.0404, -0.2412, -0.2569],\n",
              "           [ 0.0956, -0.1649,  0.2583]]],\n",
              " \n",
              " \n",
              "         [[[-0.1102,  0.0973,  0.2796],\n",
              "           [ 0.0233,  0.1389,  0.2754],\n",
              "           [-0.0719,  0.0912, -0.2213]]],\n",
              " \n",
              " \n",
              "         [[[ 0.1372, -0.1271, -0.1887],\n",
              "           [ 0.1699, -0.3249, -0.0157],\n",
              "           [ 0.0005,  0.2904, -0.0377]]],\n",
              " \n",
              " \n",
              "         [[[-0.0623,  0.1991,  0.1723],\n",
              "           [ 0.1764, -0.3121, -0.0612],\n",
              "           [ 0.2015, -0.1517, -0.0890]]],\n",
              " \n",
              " \n",
              "         [[[ 0.0057, -0.2032, -0.2188],\n",
              "           [ 0.0152,  0.1251,  0.2836],\n",
              "           [-0.0227,  0.1915,  0.3145]]],\n",
              " \n",
              " \n",
              "         [[[ 0.0322, -0.2940, -0.2321],\n",
              "           [-0.0919,  0.0318, -0.0626],\n",
              "           [ 0.3265, -0.0182,  0.1224]]],\n",
              " \n",
              " \n",
              "         [[[ 0.0329, -0.2783, -0.3309],\n",
              "           [-0.3030, -0.0085, -0.0327],\n",
              "           [ 0.2087,  0.2853,  0.0730]]],\n",
              " \n",
              " \n",
              "         [[[ 0.3088, -0.2014, -0.2409],\n",
              "           [-0.2236, -0.0978,  0.0795],\n",
              "           [-0.2647, -0.0787,  0.2337]]]], requires_grad=True),\n",
              " Parameter containing:\n",
              " tensor([ 0.2665, -0.1529,  0.3087,  0.0161,  0.2111, -0.0548,  0.0955,  0.0675,\n",
              "          0.3251, -0.0602, -0.2022, -0.0257,  0.0417, -0.0917, -0.1251, -0.0662,\n",
              "         -0.0945, -0.3012,  0.0821,  0.0015, -0.1068,  0.2556,  0.1300,  0.0217,\n",
              "          0.1051,  0.0029,  0.1261,  0.2404, -0.0290, -0.2618,  0.3253, -0.0884],\n",
              "        requires_grad=True),\n",
              " Parameter containing:\n",
              " tensor([[[[-0.0372, -0.0159,  0.0495],\n",
              "           [-0.0086, -0.0426,  0.0289],\n",
              "           [ 0.0476,  0.0273, -0.0338]],\n",
              " \n",
              "          [[-0.0200,  0.0307,  0.0207],\n",
              "           [-0.0136, -0.0408,  0.0539],\n",
              "           [-0.0565,  0.0066,  0.0148]],\n",
              " \n",
              "          [[ 0.0183, -0.0246,  0.0429],\n",
              "           [ 0.0182, -0.0194, -0.0399],\n",
              "           [ 0.0208, -0.0352, -0.0194]],\n",
              " \n",
              "          ...,\n",
              " \n",
              "          [[-0.0332,  0.0150,  0.0387],\n",
              "           [ 0.0508,  0.0153, -0.0314],\n",
              "           [ 0.0493, -0.0141,  0.0108]],\n",
              " \n",
              "          [[ 0.0174,  0.0316,  0.0371],\n",
              "           [-0.0475,  0.0349, -0.0255],\n",
              "           [-0.0163, -0.0273,  0.0450]],\n",
              " \n",
              "          [[ 0.0336, -0.0165,  0.0184],\n",
              "           [-0.0159,  0.0350,  0.0217],\n",
              "           [-0.0460,  0.0015, -0.0434]]],\n",
              " \n",
              " \n",
              "         [[[-0.0374,  0.0007,  0.0329],\n",
              "           [-0.0446,  0.0106,  0.0380],\n",
              "           [-0.0096,  0.0374, -0.0319]],\n",
              " \n",
              "          [[-0.0518,  0.0553, -0.0260],\n",
              "           [-0.0098,  0.0444, -0.0317],\n",
              "           [-0.0343,  0.0391,  0.0455]],\n",
              " \n",
              "          [[-0.0541, -0.0064, -0.0571],\n",
              "           [-0.0124, -0.0428, -0.0371],\n",
              "           [-0.0389, -0.0067, -0.0145]],\n",
              " \n",
              "          ...,\n",
              " \n",
              "          [[ 0.0517,  0.0345,  0.0108],\n",
              "           [-0.0135, -0.0035, -0.0429],\n",
              "           [ 0.0588,  0.0447,  0.0374]],\n",
              " \n",
              "          [[ 0.0340,  0.0145, -0.0456],\n",
              "           [-0.0397, -0.0290,  0.0104],\n",
              "           [ 0.0518, -0.0275, -0.0268]],\n",
              " \n",
              "          [[-0.0521, -0.0034,  0.0042],\n",
              "           [ 0.0091,  0.0164, -0.0453],\n",
              "           [ 0.0395,  0.0126,  0.0219]]],\n",
              " \n",
              " \n",
              "         [[[ 0.0061, -0.0032,  0.0265],\n",
              "           [-0.0555,  0.0305, -0.0329],\n",
              "           [ 0.0509,  0.0021, -0.0541]],\n",
              " \n",
              "          [[ 0.0459, -0.0252, -0.0296],\n",
              "           [ 0.0205,  0.0022, -0.0455],\n",
              "           [-0.0444,  0.0580,  0.0430]],\n",
              " \n",
              "          [[-0.0176, -0.0562, -0.0228],\n",
              "           [-0.0260, -0.0066, -0.0542],\n",
              "           [-0.0140,  0.0582,  0.0323]],\n",
              " \n",
              "          ...,\n",
              " \n",
              "          [[ 0.0426, -0.0128,  0.0003],\n",
              "           [ 0.0073,  0.0546,  0.0469],\n",
              "           [ 0.0193,  0.0302, -0.0467]],\n",
              " \n",
              "          [[-0.0137, -0.0179, -0.0522],\n",
              "           [-0.0056, -0.0283, -0.0156],\n",
              "           [-0.0287, -0.0509,  0.0270]],\n",
              " \n",
              "          [[ 0.0558, -0.0105,  0.0171],\n",
              "           [ 0.0100,  0.0417, -0.0323],\n",
              "           [ 0.0224,  0.0322,  0.0089]]],\n",
              " \n",
              " \n",
              "         ...,\n",
              " \n",
              " \n",
              "         [[[ 0.0221,  0.0587, -0.0183],\n",
              "           [ 0.0076, -0.0332,  0.0002],\n",
              "           [-0.0407,  0.0101, -0.0225]],\n",
              " \n",
              "          [[ 0.0273, -0.0347,  0.0211],\n",
              "           [-0.0495, -0.0555,  0.0406],\n",
              "           [ 0.0489,  0.0346, -0.0361]],\n",
              " \n",
              "          [[ 0.0103,  0.0126, -0.0276],\n",
              "           [ 0.0333,  0.0366, -0.0404],\n",
              "           [-0.0531,  0.0107, -0.0175]],\n",
              " \n",
              "          ...,\n",
              " \n",
              "          [[-0.0450,  0.0513, -0.0417],\n",
              "           [-0.0389, -0.0354, -0.0201],\n",
              "           [ 0.0412, -0.0003,  0.0504]],\n",
              " \n",
              "          [[-0.0125, -0.0229, -0.0574],\n",
              "           [-0.0027, -0.0585, -0.0022],\n",
              "           [-0.0329,  0.0077, -0.0441]],\n",
              " \n",
              "          [[-0.0352, -0.0004,  0.0153],\n",
              "           [-0.0337, -0.0526,  0.0456],\n",
              "           [ 0.0166,  0.0575, -0.0312]]],\n",
              " \n",
              " \n",
              "         [[[-0.0154,  0.0576, -0.0281],\n",
              "           [-0.0116,  0.0258,  0.0347],\n",
              "           [ 0.0515,  0.0548,  0.0409]],\n",
              " \n",
              "          [[-0.0376, -0.0395, -0.0227],\n",
              "           [ 0.0138,  0.0363,  0.0550],\n",
              "           [-0.0120, -0.0066,  0.0137]],\n",
              " \n",
              "          [[ 0.0551,  0.0260,  0.0258],\n",
              "           [ 0.0213, -0.0520, -0.0369],\n",
              "           [-0.0162, -0.0006, -0.0018]],\n",
              " \n",
              "          ...,\n",
              " \n",
              "          [[ 0.0386,  0.0291,  0.0379],\n",
              "           [-0.0041, -0.0557, -0.0016],\n",
              "           [-0.0477, -0.0586, -0.0243]],\n",
              " \n",
              "          [[-0.0285, -0.0126, -0.0030],\n",
              "           [-0.0048,  0.0107,  0.0361],\n",
              "           [ 0.0392,  0.0455,  0.0448]],\n",
              " \n",
              "          [[ 0.0140, -0.0308,  0.0312],\n",
              "           [ 0.0500,  0.0527, -0.0344],\n",
              "           [-0.0064, -0.0402,  0.0327]]],\n",
              " \n",
              " \n",
              "         [[[-0.0537,  0.0260, -0.0561],\n",
              "           [-0.0056,  0.0454, -0.0042],\n",
              "           [-0.0251, -0.0563,  0.0073]],\n",
              " \n",
              "          [[-0.0004, -0.0046, -0.0055],\n",
              "           [ 0.0048,  0.0183, -0.0487],\n",
              "           [-0.0173,  0.0278,  0.0153]],\n",
              " \n",
              "          [[ 0.0198, -0.0218,  0.0452],\n",
              "           [-0.0269, -0.0281, -0.0069],\n",
              "           [ 0.0199,  0.0490,  0.0213]],\n",
              " \n",
              "          ...,\n",
              " \n",
              "          [[ 0.0190,  0.0245, -0.0381],\n",
              "           [ 0.0149, -0.0180, -0.0425],\n",
              "           [ 0.0575,  0.0231,  0.0327]],\n",
              " \n",
              "          [[ 0.0318, -0.0545,  0.0308],\n",
              "           [-0.0568,  0.0180, -0.0286],\n",
              "           [ 0.0221,  0.0432, -0.0378]],\n",
              " \n",
              "          [[-0.0050,  0.0371, -0.0485],\n",
              "           [-0.0078,  0.0055,  0.0305],\n",
              "           [ 0.0068, -0.0066,  0.0064]]]], requires_grad=True),\n",
              " Parameter containing:\n",
              " tensor([-0.0360,  0.0329, -0.0243,  0.0521, -0.0011,  0.0242,  0.0007,  0.0043,\n",
              "         -0.0380,  0.0412,  0.0006, -0.0188,  0.0539,  0.0464,  0.0573,  0.0113,\n",
              "         -0.0570,  0.0471, -0.0521,  0.0462,  0.0444, -0.0375, -0.0391,  0.0139,\n",
              "         -0.0071, -0.0518,  0.0178,  0.0218, -0.0538, -0.0138, -0.0265, -0.0081,\n",
              "          0.0113,  0.0011, -0.0447, -0.0516,  0.0259, -0.0456,  0.0127, -0.0094,\n",
              "          0.0187, -0.0136,  0.0048, -0.0003,  0.0100, -0.0283,  0.0581, -0.0513,\n",
              "         -0.0079,  0.0201,  0.0148, -0.0368,  0.0168, -0.0198,  0.0281, -0.0480,\n",
              "          0.0324,  0.0293,  0.0178,  0.0517,  0.0445,  0.0257, -0.0334, -0.0383],\n",
              "        requires_grad=True),\n",
              " Parameter containing:\n",
              " tensor([[[[-0.0130, -0.0435],\n",
              "           [-0.0314,  0.0537]],\n",
              " \n",
              "          [[-0.0524, -0.0223],\n",
              "           [ 0.0170,  0.0857]],\n",
              " \n",
              "          [[-0.0526, -0.0633],\n",
              "           [-0.0505,  0.0387]],\n",
              " \n",
              "          ...,\n",
              " \n",
              "          [[-0.0377, -0.0108],\n",
              "           [-0.0084,  0.0225]],\n",
              " \n",
              "          [[-0.0087,  0.0859],\n",
              "           [ 0.0074,  0.0763]],\n",
              " \n",
              "          [[ 0.0191,  0.0450],\n",
              "           [ 0.0557,  0.0749]]],\n",
              " \n",
              " \n",
              "         [[[ 0.0151,  0.0617],\n",
              "           [ 0.0491,  0.0296]],\n",
              " \n",
              "          [[ 0.0793, -0.0330],\n",
              "           [-0.0021,  0.0478]],\n",
              " \n",
              "          [[-0.0805, -0.0283],\n",
              "           [ 0.0155,  0.0776]],\n",
              " \n",
              "          ...,\n",
              " \n",
              "          [[-0.0646,  0.0743],\n",
              "           [ 0.0403,  0.0062]],\n",
              " \n",
              "          [[ 0.0464,  0.0582],\n",
              "           [ 0.0037, -0.0157]],\n",
              " \n",
              "          [[ 0.0654,  0.0511],\n",
              "           [ 0.0860, -0.0799]]],\n",
              " \n",
              " \n",
              "         [[[ 0.0697, -0.0501],\n",
              "           [-0.0555, -0.0115]],\n",
              " \n",
              "          [[-0.0328, -0.0516],\n",
              "           [-0.0665,  0.0302]],\n",
              " \n",
              "          [[ 0.0158, -0.0209],\n",
              "           [ 0.0072,  0.0702]],\n",
              " \n",
              "          ...,\n",
              " \n",
              "          [[-0.0271,  0.0156],\n",
              "           [-0.0300, -0.0124]],\n",
              " \n",
              "          [[-0.0864,  0.0576],\n",
              "           [-0.0399,  0.0326]],\n",
              " \n",
              "          [[ 0.0773, -0.0581],\n",
              "           [-0.0665,  0.0278]]],\n",
              " \n",
              " \n",
              "         ...,\n",
              " \n",
              " \n",
              "         [[[ 0.0806, -0.0475],\n",
              "           [-0.0660,  0.0722]],\n",
              " \n",
              "          [[-0.0143, -0.0876],\n",
              "           [ 0.0064,  0.0599]],\n",
              " \n",
              "          [[ 0.0022, -0.0560],\n",
              "           [-0.0094,  0.0639]],\n",
              " \n",
              "          ...,\n",
              " \n",
              "          [[ 0.0587,  0.0447],\n",
              "           [ 0.0297, -0.0846]],\n",
              " \n",
              "          [[-0.0489, -0.0213],\n",
              "           [-0.0697,  0.0601]],\n",
              " \n",
              "          [[-0.0446,  0.0133],\n",
              "           [ 0.0267, -0.0248]]],\n",
              " \n",
              " \n",
              "         [[[-0.0040, -0.0088],\n",
              "           [-0.0034,  0.0803]],\n",
              " \n",
              "          [[ 0.0704, -0.0588],\n",
              "           [ 0.0711, -0.0256]],\n",
              " \n",
              "          [[ 0.0761,  0.0603],\n",
              "           [-0.0498, -0.0455]],\n",
              " \n",
              "          ...,\n",
              " \n",
              "          [[-0.0361,  0.0207],\n",
              "           [-0.0840,  0.0857]],\n",
              " \n",
              "          [[-0.0691, -0.0049],\n",
              "           [ 0.0336,  0.0033]],\n",
              " \n",
              "          [[-0.0428,  0.0564],\n",
              "           [-0.0384, -0.0021]]],\n",
              " \n",
              " \n",
              "         [[[-0.0081, -0.0400],\n",
              "           [-0.0812, -0.0486]],\n",
              " \n",
              "          [[ 0.0701, -0.0152],\n",
              "           [-0.0808,  0.0827]],\n",
              " \n",
              "          [[ 0.0818,  0.0133],\n",
              "           [ 0.0551, -0.0745]],\n",
              " \n",
              "          ...,\n",
              " \n",
              "          [[-0.0012,  0.0432],\n",
              "           [-0.0144,  0.0244]],\n",
              " \n",
              "          [[-0.0272,  0.0252],\n",
              "           [-0.0773, -0.0296]],\n",
              " \n",
              "          [[-0.0271, -0.0665],\n",
              "           [-0.0413, -0.0226]]]], requires_grad=True),\n",
              " Parameter containing:\n",
              " tensor([-0.0571, -0.0356,  0.0416, -0.0853, -0.0500,  0.0858, -0.0159, -0.0776,\n",
              "         -0.0824, -0.0031, -0.0527,  0.0023, -0.0552, -0.0521, -0.0421, -0.0354,\n",
              "          0.0170,  0.0203,  0.0828, -0.0539, -0.0018,  0.0779,  0.0127,  0.0504,\n",
              "         -0.0268, -0.0840, -0.0220, -0.0223, -0.0500,  0.0198,  0.0432, -0.0296],\n",
              "        requires_grad=True),\n",
              " Parameter containing:\n",
              " tensor([[[[-0.0556,  0.0891],\n",
              "           [ 0.4149,  0.2554]]],\n",
              " \n",
              " \n",
              "         [[[ 0.0974,  0.4095],\n",
              "           [-0.3295, -0.3099]]],\n",
              " \n",
              " \n",
              "         [[[-0.0849,  0.0217],\n",
              "           [ 0.0643,  0.3686]]],\n",
              " \n",
              " \n",
              "         [[[-0.4545,  0.3040],\n",
              "           [ 0.1455,  0.3952]]],\n",
              " \n",
              " \n",
              "         [[[ 0.3354,  0.0385],\n",
              "           [ 0.4182,  0.4989]]],\n",
              " \n",
              " \n",
              "         [[[ 0.1102,  0.4123],\n",
              "           [-0.0293, -0.4941]]],\n",
              " \n",
              " \n",
              "         [[[-0.2827,  0.4339],\n",
              "           [ 0.3831,  0.2364]]],\n",
              " \n",
              " \n",
              "         [[[ 0.1498, -0.2613],\n",
              "           [ 0.0602, -0.4118]]],\n",
              " \n",
              " \n",
              "         [[[ 0.4139,  0.2181],\n",
              "           [-0.3229,  0.2905]]],\n",
              " \n",
              " \n",
              "         [[[ 0.2360, -0.2608],\n",
              "           [-0.1554,  0.4142]]],\n",
              " \n",
              " \n",
              "         [[[-0.1948, -0.0987],\n",
              "           [ 0.2857, -0.0077]]],\n",
              " \n",
              " \n",
              "         [[[-0.0335,  0.4040],\n",
              "           [-0.3040,  0.4607]]],\n",
              " \n",
              " \n",
              "         [[[ 0.4843,  0.1147],\n",
              "           [ 0.3878,  0.3381]]],\n",
              " \n",
              " \n",
              "         [[[-0.2157,  0.1347],\n",
              "           [-0.3425,  0.3917]]],\n",
              " \n",
              " \n",
              "         [[[-0.0399, -0.3906],\n",
              "           [ 0.4913,  0.4859]]],\n",
              " \n",
              " \n",
              "         [[[ 0.4023, -0.4846],\n",
              "           [-0.1835,  0.3681]]],\n",
              " \n",
              " \n",
              "         [[[-0.3619, -0.0819],\n",
              "           [-0.2532, -0.1537]]],\n",
              " \n",
              " \n",
              "         [[[-0.2126, -0.3478],\n",
              "           [ 0.3488, -0.1635]]],\n",
              " \n",
              " \n",
              "         [[[ 0.3560, -0.4938],\n",
              "           [-0.1881,  0.1845]]],\n",
              " \n",
              " \n",
              "         [[[-0.0058, -0.2191],\n",
              "           [ 0.4632, -0.1191]]],\n",
              " \n",
              " \n",
              "         [[[ 0.1568, -0.1745],\n",
              "           [ 0.2172,  0.3811]]],\n",
              " \n",
              " \n",
              "         [[[ 0.4276, -0.2891],\n",
              "           [ 0.2614,  0.0933]]],\n",
              " \n",
              " \n",
              "         [[[-0.4446, -0.3512],\n",
              "           [ 0.0941,  0.3148]]],\n",
              " \n",
              " \n",
              "         [[[ 0.3044, -0.2386],\n",
              "           [ 0.0216, -0.1461]]],\n",
              " \n",
              " \n",
              "         [[[-0.1780, -0.1515],\n",
              "           [ 0.2280,  0.0331]]],\n",
              " \n",
              " \n",
              "         [[[ 0.3274, -0.4733],\n",
              "           [-0.1093, -0.0103]]],\n",
              " \n",
              " \n",
              "         [[[-0.3588,  0.0063],\n",
              "           [ 0.2858, -0.4690]]],\n",
              " \n",
              " \n",
              "         [[[ 0.3670,  0.4292],\n",
              "           [ 0.2516, -0.0752]]],\n",
              " \n",
              " \n",
              "         [[[-0.4516,  0.3922],\n",
              "           [-0.4784, -0.2938]]],\n",
              " \n",
              " \n",
              "         [[[ 0.2837,  0.1717],\n",
              "           [-0.3666, -0.4806]]],\n",
              " \n",
              " \n",
              "         [[[ 0.0088,  0.1040],\n",
              "           [-0.3695, -0.1912]]],\n",
              " \n",
              " \n",
              "         [[[-0.0309, -0.1075],\n",
              "           [ 0.3001, -0.2352]]]], requires_grad=True),\n",
              " Parameter containing:\n",
              " tensor([0.3816], requires_grad=True)]"
            ]
          },
          "metadata": {},
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "opt = Adam(model.parameters())\n",
        "max_epoch = 10\n",
        "model = model.to(device)\n",
        "\n",
        "for epoch in range(max_epoch):\n",
        "  training_loss = 0.0\n",
        "\n",
        "  for data, target in train_loader:\n",
        "    data = data.to(device)\n",
        "    target = target.to(device)\n",
        "\n",
        "    opt.zero_grad()\n",
        "\n",
        "    output = model(data)\n",
        "\n",
        "    loss = F.mse_loss(output, target)\n",
        "    loss.backward()\n",
        "\n",
        "    opt.step()\n",
        "\n",
        "    training_loss += loss.item()\n",
        "\n",
        "  if (epoch+1)%2 == 0:\n",
        "    print(\"epoch: {} training loss : {}\".format(epoch+1,training_loss))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "QpY4QlJKjkHL",
        "outputId": "e5aebd04-f2e5-4dc0-8744-38d7458dcf65"
      },
      "execution_count": 19,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch: 2 training loss : 1.966891153715551\n",
            "epoch: 4 training loss : 1.391926234588027\n",
            "epoch: 6 training loss : 1.1925505688413978\n",
            "epoch: 8 training loss : 1.0904792752116919\n",
            "epoch: 10 training loss : 1.026395377703011\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "x_test, _ = torch.load(\"test.pt\")"
      ],
      "metadata": {
        "id": "Weezihc4jqUF"
      },
      "execution_count": 20,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "img_id = 100\n",
        "test1 = x_test[img_id]\n",
        "test1_noise = salt_n_pepper_noise(normalize(test1.reshape(-1,1,28,28))) #here we are supplyinhy noised images as input (test data)\n",
        "test1_output = model(test1_noise.float().to(device)) #denoised image"
      ],
      "metadata": {
        "id": "uVROFgyiotpA"
      },
      "execution_count": 21,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "plt.imshow(test1_output.cpu().detach().numpy().reshape(28,28)) #denoised image , so noise is quite efficiently removed in this image"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "id": "OFBt77MfpNDD",
        "outputId": "d20fd800-5151-43f9-e434-9bce50685785"
      },
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f2cbd90b950>"
            ]
          },
          "metadata": {},
          "execution_count": 22
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAD4CAYAAAAq5pAIAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAWZklEQVR4nO3dbYxc5XUH8P+ZWe+bvX43xm+8GZfEodSQDU2ANk7zIqBtTBqJhg8plWicD6EKUioV0UrhI6qapHyoUjkBxVSUlDQgUEJTHIqC0grCQgyYd2rWxmbtxTZev+7O7s7phx2iBfb5n2Xu7My0z/8nWbueZ++9z9yZs3d3zz3nMXeHiPz/V2r1BESkORTsIplQsItkQsEukgkFu0gmOpp5sM6OXu/pXJz+gigzwMZLwfetaN9mxbYvsu9IgWN7KTh2wedtRZI5RV7v2WDPrejrHU2t4EvO953e+enKUVQmTs34BYWC3cyuBHA7gDKA77v7bezrezoX4+Pn35DeX7XKDzg+kRzyrk66qU1O0nGfVw62J6/uRLDvruA0B087mjvddS8/L9VgbqWx9DkHABsP5kaem5HXczbjEe9OP3cPLg4WBXs0t3nRa85ODP9Owd6rj7/8/eRY3T/Gm1kZwD8CuArARgDXmdnGevcnInOryO/slwJ4zd13u3sFwA8BbGnMtESk0YoE+xoAb0z7/77aY+9iZlvNbMDMBioTJwscTkSKmPO/xrv7Nnfvd/f+zo75c304EUkoEuz7Aayb9v+1tcdEpA0VCfYnAWwws3PNrBPAlwA82JhpiUij1Z16c/cJM7sRwH9gKvV2p7s/H2wUpzTY5iS9FqZpOoLUWpRCYqmaMv+e6eXg2ODHrkZpRTL36HmVgjRPkdQaEKRTg9fEw/HgWkXSpWE6Mzgv1b5uOl46Pc73z1SCbaN7ShIK5dnd/SEADxXZh4g0h26XFcmEgl0kEwp2kUwo2EUyoWAXyYSCXSQTTa1nhxm8QI0xzdEHue6wPrnO3CWAYvPGLPLwE0Eymzh2wSI6fmQjP/ZZ/36cjtt4VJZMto1ek6Dk2YpVwBZSGuUHj0qL2WtqUa19ub5ieV3ZRTKhYBfJhIJdJBMKdpFMKNhFMqFgF8lEc1Nv1SpsrJIejzpyMlF6K+oeG3RRpeWYUXltkEopRV11A6zU88Dl/NjdB/m+y0dO8GPP56We9PWOUkxBiWv0mle75iXHSqfG+L6jXtFBqjc6b9UF6fMWdr5lcyfvJV3ZRTKhYBfJhIJdJBMKdpFMKNhFMqFgF8mEgl0kE83Ns8NobjXKhVMFl2T27nRONtq/RV2Do2+pQZnoxDK+ks7gH/cmx+yM03Tb0pvpbQGgupgfO2rnzEajNtVRvnlyUdDOmd07UbRNdfR+C+4ZKVK2zN/L6TFd2UUyoWAXyYSCXSQTCnaRTCjYRTKhYBfJhIJdJBPNzbOXDN6ZzmcXyT3aqVE67r08JxtiedWolTRZOhgAvIu/DK9/nufC//CzTybHfrrjY3TbZbv4TQLloSN0vLqct6pm5y2s2w7q/DuGj/FDk6WuaUtzAFYJehRUi9W784MHdf60HXt6qFCwm9kggOMAJgFMuHt/kf2JyNxpxJX9U+5+qAH7EZE5pN/ZRTJRNNgdwMNm9pSZbZ3pC8xsq5kNmNlAZeJkwcOJSL2K/hh/hbvvN7MzAOwws5fc/bHpX+Du2wBsA4BFvauDv2qIyFwpdGV39/21j8MA7gdwaSMmJSKNV3ewm9l8M+t753MAnwOwq1ETE5HGKvJj/EoA99eWl+0A8C/u/jO6hTtsktcw18t7uvgXsP7lALybbx/lfKkJ/pxPnr+Ujt/0+Z/Q8UPjfcmxc37K7z8on+Z5dl+c3jcQ3xvhdS4vDMR5eJA8OgAY6+cf5MHDZbRLwW+kwWtOjx9tW6e6g93ddwP4nQbORUTmkFJvIplQsItkQsEukgkFu0gmFOwimWhyK2nw8rwo5cC2jVr/RmmaKCXIShqjpYUPvU2HTy9bQcfL4OmtHUMfSo4t2sdLVD1qqRyct9JJ3qoapKQ5Ku1FVAJ7OkinknbOFrWCjkSp2Oj9VKBkGqX60pm6sotkQsEukgkFu0gmFOwimVCwi2RCwS6SCQW7SCaam2d30Fw6azMNBLnwKO8Z5fCjnC4rkR0PvmcGedNJnsrGT4YvouNDL52RHOvrO0q3jZZNLgUtusHKSAGeZw+W6C4dD459hD83rE7fv1CN7h8YOcX3HSzJjKBElgpy9FYh73VyP4iu7CKZULCLZELBLpIJBbtIJhTsIplQsItkQsEukonm5tkNND9ZpKY8qo22aIndQNiqmqiczVtFj1zB88m/NW+Mjp/5X+nnFi4X3cnP28QSvlx0qRLk6UmuPMrxj561mI7PWxzMbSx9D0CUR/ee4OaHQNh6nN3XEbW5ZvdtkFp3XdlFMqFgF8mEgl0kEwp2kUwo2EUyoWAXyYSCXSQTze8bz/KP0RK9pH96lE+mfboxi1r6UZLrDmqbJ3p5bfNNFz9Mx2/f+Qd0fP2PnkgPfuQCum2kY4TfA1Dt5fnokxvS9xjsvYbnojdf+DIdf/PkIjq+5z/PSo7Nf5O/H7pH+NwW7D5Ox+14kMdfOD89OBnk6KN1ChLCK7uZ3Wlmw2a2a9pjS81sh5m9Wvu4pK6ji0jTzObH+B8AuPI9j90M4BF33wDgkdr/RaSNhcHu7o8BeO8aQlsAbK99vh3ANQ2el4g0WL1/oFvp7kO1zw8AWJn6QjPbamYDZjZQmQz6eonInCn813h3d0y1kkyNb3P3fnfv7yzzwgURmTv1BvtBM1sFALWPw42bkojMhXqD/UEA19c+vx7AA42ZjojMlTDPbmb3ANgMYLmZ7QPwTQC3AbjXzG4AsAfAtbM+IqvFjWqAC2xrlXG+72i9bnIPgAc9wg9/mOfwf6/3FTr+D8NX0fHyCtIfPahXL1WCvu/BeZmYz5/b3i+ma9a3f/IOuu0nuni9+zzj533neel7I/5tpJ9ue/fjn6Djqx5dSMcXBmsF0Hr3aA2DqO9DQhjs7n5dYujTdR1RRFpCt8uKZELBLpIJBbtIJhTsIplQsItkoslLNjtsNL30cbWvh25up8m2C/m26OalmFE6w7vSKabKoqDM87d5mej583h6q+/1IBXTly6XtCilGCy5/PZH02k9APjTv/0ZHR8cXZ4cu+WVP6HbHnwmeRc2AKB3iKe3KqQCdnQVf95fuewXdPz7vZfT8eg1Kx0ht45Hyz2z9yp5vXVlF8mEgl0kEwp2kUwo2EUyoWAXyYSCXSQTCnaRTDQ3z14qodrbnRxmeXQAtIy1dOw03dS7g1bRwfLBNpouke0ePEC37fjM+XR86x5ewtr3Bp/bxPK+5BhbthgAqn3p1wMADm/hrcT+YtFLdPyP/uovk2MLXztMt+2bP0LHSwf49k5aMo9vXEu3ffoj6+g4qjzHXyZLVQOAz0+fdw9KXEunSFtzUlqrK7tIJhTsIplQsItkQsEukgkFu0gmFOwimVCwi2Si+fXsrH46XIqWjEete4N8swfLLhvLfa7gi9iee98JOv6rDr6s8tkjvA32sfXpeva+vTzfO3QZX6Vn28f+iY5/auef0fEVg0eTY94V9BgI2lxPrkrXygNAaSx93sYX8Nd7bW963gCwN3jNoz4BmJd+L4ftvcv1XaN1ZRfJhIJdJBMKdpFMKNhFMqFgF8mEgl0kEwp2kUw0N88OACWSDw9yk96Tzssaq/EFgM6gnr1If/UgR19+nde7d769gY6XKryefZLkq6NtT6zn53x4Ml0rDwCVn/Ncd7U3XZMe9RBAsIJ3eeQkHfeRY8mx8QuX0m0f27+ejh87wdcpWNbL72+InhtD692L1LOb2Z1mNmxmu6Y9dquZ7TeznbV/V3/A+YpIk83mx/gfALhyhse/4+6bav8eauy0RKTRwmB398cAHGnCXERkDhX5A92NZvZs7cf85I3CZrbVzAbMbKAyyfuZicjcqTfYvwtgPYBNAIYAfCv1he6+zd373b2/s8yLLkRk7tQV7O5+0N0n3b0K4HsALm3stESk0eoKdjNbNe2/XwCwK/W1ItIewjy7md0DYDOA5Wa2D8A3AWw2s00AHMAggK/O6mhmcLL2NK9IB0on0rnLqB494kE9fInl2Sd4vnjsQt6DfLKb5/gPbeK//iwcTM+tdIr34u/Zz/PoT544j45XFtJh2ARJKEe55gqv468u4D3v2Ws2ch6/zn3xbH79uuu/+frso6v4+7FrmKxzEK1hQNZPYOuzhxHi7tfN8PAd0XYi0l50u6xIJhTsIplQsItkQsEukgkFu0gmmt9KmpWSBukvll4LU2/BtzWb5Omvah8paQxSSEfP76LjlRVBqWeJt9ge2ZB+cr1Dy+i2oxt4KebRcZ7263krKA1mr3eZv97VxekW2QDgrFwawNgFK5NjJy/gJdGjVV4S3X0gSK0d4stN22T6NbdRPjerknNKzreu7CKZULCLZELBLpIJBbtIJhTsIplQsItkQsEukonmt5JmonbOJK8atoI+zcslo3bQNpHev3fw75lLn+ftuMqjPJf91u/yRP5ll7ycHPv10Bq67RWr36DjVy55jo7/Yt1FdLy6Od2yuTTOX7O+ffz+g9NL+f0HJ85Kv196+tJtpgHg3sd5P5aVr/HXpHSMlLAC/P0W3m9S39LlurKLZELBLpIJBbtIJhTsIplQsItkQsEukgkFu0gmml7PTtsDRzXpLJfO2usCqC7kS+xG9exOaq9p3hNAxwivGV/2FM/Jji3myyI/feba5Nh4hZ/Tgf1n0fFLFu6l44suOkzHD69Ot6oudfI8+tvH0ktRA8CZZw/T8dN70rX8C37Fe2CfNxDUlDtf6tomgyYHE+n9h70ZWByQIV3ZRTKhYBfJhIJdJBMKdpFMKNhFMqFgF8mEgl0kE83Ns5sBHemcdFQXzpb/jXLdpVGeF612BqeC1AmHOfpgbtH44t28Fn/s8ILk2IHP8Of94TUH6Pgzx/ly05tXv0rHP7R+KDk27vx5/+jNj9LxwV2r6fiGf03fv9BxmN8fEN3zES3x7d38HgEvpd/rxpYHD8cL9I03s3Vm9qiZvWBmz5vZ12uPLzWzHWb2au3jkmhfItI6s/kxfgLAN9x9I4CPA/iamW0EcDOAR9x9A4BHav8XkTYVBru7D7n707XPjwN4EcAaAFsAbK992XYA18zVJEWkuA/0BzozOwfAxQCeALDS3d/5hewAgBkX1jKzrWY2YGYDlUnei01E5s6sg93MFgD4MYCb3P1d3frc3ZH4y4C7b3P3fnfv7yzzxooiMndmFexmNg9TgX63u99Xe/igma2qja8CwEuQRKSlwtSbmRmAOwC86O7fnjb0IIDrAdxW+/hAeDQzeFd6KVwbD5YuZruu8HRFmJoLtqeCVIl38uV/S8Hz7t3NS2S756eXhO4a4aW9+9eeS8d3beRpxfIKPrfJiYuTYx37+VLWfYN0GOuf5b8WdoyQ0uGojDRgYxX+BVFrcrqUNb8GV7u604MkpTebZ3w5gC8DeM7MdtYeuwVTQX6vmd0AYA+Aa2exLxFpkTDY3f2XAFJ3EHy6sdMRkbmi22VFMqFgF8mEgl0kEwp2kUwo2EUy0fwlm0mZKiZ4vjlsscsEnX2jkkUqyKNb0OY6Ku1FR5CzJfcI9Ow5TrftPMJz3ct/HbTo7uFzY220fR4v3Y3unYjuy7DRdC68uoDkqoHw/WKkVBtA+F6mufSgDTXN0ZMxXdlFMqFgF8mEgl0kEwp2kUwo2EUyoWAXyYSCXSQTzc+zk28v3lWgnfMoz9lG+2ZtqoGg9W+URyfbAnEraowHz60n3bZ4so+3NE7WM85S594jdJzdG2FB3TarzQbi8wrWO6HA6z21g+DEFRkveF6Sm9W1lYj8n6NgF8mEgl0kEwp2kUwo2EUyoWAXyYSCXSQTzc2zu/Ma5CivykqIo9xkUJ8c1hBXSK67zGubrUitPEBrlAHATo0lxzoK9OIHULzHABn3Mj8v0f0HpcNH6bgvTC9lHV3mbLLYeWP3PgAAgnsz6L7JGgheSp9TXdlFMqFgF8mEgl0kEwp2kUwo2EUyoWAXyYSCXSQTs1mffR2AuwCsBOAAtrn77WZ2K4CvAHir9qW3uPtD4RFZLj3IJ6NK8rLBGukW9HYP88WsHj7qMR48r6hnvS/ga6yz/RdZ8x6I88Xh/kk+2bxYTfjkmcvoeGmM3BsR7Lvayd8PpeN8bXhWSw8ATu7NoPMOGHmrzeammgkA33D3p82sD8BTZrajNvYdd//7umcmIk0zm/XZhwAM1T4/bmYvAlgz1xMTkcb6QL+zm9k5AC4G8ETtoRvN7Fkzu9PMliS22WpmA2Y2UJkMfvQRkTkz62A3swUAfgzgJnc/BuC7ANYD2ISpK/+3ZtrO3be5e7+793eWexswZRGpx6yC3czmYSrQ73b3+wDA3Q+6+6S7VwF8D8ClczdNESkqDHabKtm6A8CL7v7taY+vmvZlXwCwq/HTE5FGmc1f4y8H8GUAz5nZztpjtwC4zsw2YSodNwjgq7M6IktDBekQmsIqspwzEH/bI+m1sBwyKnElJYuz2j+ZW5TWi9KCUcvlUJXsPzrnQXltKUrVsrLlYMnlMKUYpXKj8l2y/3AJ7zpLpmfz1/hfYubu4nFOXUTahu6gE8mEgl0kEwp2kUwo2EUyoWAXyYSCXSQTzW0lbcaXPh6r8O1ZLj3KuUb55PEgn1wgj18Nyh2jksZwyWfWljjKyQa57LBFd7R/tn3QvtuDXDaCXLaT814a5e81Y/cHIC6JLp0K3svkNWWtogHU3YZaV3aRTCjYRTKhYBfJhIJdJBMKdpFMKNhFMqFgF8mEeZSfbuTBzN4CsGfaQ8sBHGraBD6Ydp1bu84L0Nzq1ci5ne3uK2YaaGqwv+/gZgPu3t+yCRDtOrd2nRegudWrWXPTj/EimVCwi2Si1cG+rcXHZ9p1bu06L0Bzq1dT5tbS39lFpHlafWUXkSZRsItkoiXBbmZXmtnLZvaamd3cijmkmNmgmT1nZjvNbKDFc7nTzIbNbNe0x5aa2Q4ze7X2ccY19lo0t1vNbH/t3O00s6tbNLd1Zvaomb1gZs+b2ddrj7f03JF5NeW8Nf13djMrA3gFwGcB7APwJIDr3P2Fpk4kwcwGAfS7e8tvwDCz3wdwAsBd7n5h7bG/A3DE3W+rfaNc4u5/3SZzuxXAiVYv411brWjV9GXGAVwD4M/RwnNH5nUtmnDeWnFlvxTAa+6+290rAH4IYEsL5tH23P0xAEfe8/AWANtrn2/H1Jul6RJzawvuPuTuT9c+Pw7gnWXGW3ruyLyaohXBvgbAG9P+vw/ttd67A3jYzJ4ys62tnswMVrr7UO3zAwBWtnIyMwiX8W6m9ywz3jbnrp7lz4vSH+je7wp3vwTAVQC+VvtxtS351O9g7ZQ7ndUy3s0ywzLjv9HKc1fv8udFtSLY9wNYN+3/a2uPtQV331/7OAzgfrTfUtQH31lBt/ZxuMXz+Y12WsZ7pmXG0QbnrpXLn7ci2J8EsMHMzjWzTgBfAvBgC+bxPmY2v/aHE5jZfACfQ/stRf0ggOtrn18P4IEWzuVd2mUZ79Qy42jxuWv58ufu3vR/AK7G1F/k/wfA37RiDol5nQfgmdq/51s9NwD3YOrHunFM/W3jBgDLADwC4FUAPwewtI3m9s8AngPwLKYCa1WL5nYFpn5EfxbAztq/q1t97si8mnLedLusSCb0BzqRTCjYRTKhYBfJhIJdJBMKdpFMKNhFMqFgF8nE/wJxETa6JkZZ5wAAAABJRU5ErkJggg==\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plt.imshow(x_test[100])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "id": "8fopq3uQpo-h",
        "outputId": "fe4c7df9-f359-49cb-aacc-560e27cf49f6"
      },
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f2cbcec76d0>"
            ]
          },
          "metadata": {},
          "execution_count": 23
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plt.imshow(test1_noise.reshape(28,28)) #original noised data"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "id": "zQy4KHmSp-yL",
        "outputId": "f9ec11c7-e42f-4ec8-f740-eb2b22a8201d"
      },
      "execution_count": 24,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f2cbd7f9dd0>"
            ]
          },
          "metadata": {},
          "execution_count": 24
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#VAE"
      ],
      "metadata": {
        "id": "pPPSNJJ292oj"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "bs = 512\n",
        "train_loader_vae = DataLoader(TensorDataset(x_train.reshape(-1,1,28,28).float(),x_train.reshape(-1,1,28,28).float()),\n",
        "                              batch_size=bs,shuffle=True)"
      ],
      "metadata": {
        "id": "wQQtZbrlqKQC"
      },
      "execution_count": 25,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "class VAE(nn.Module):\n",
        "  def __init__(self):\n",
        "    super(VAE,self).__init__()\n",
        "\n",
        "    self.enc_cv1 = nn.Conv2d(in_channels=1,out_channels=4*8,kernel_size=4,padding=1,stride=2) #14x14\n",
        "    self.enc_cv2 = nn.Conv2d(in_channels=4*8,out_channels=8*8,kernel_size=4,padding=1,stride=2) #7x7\n",
        "    self.enc_cv3 = nn.Conv2d(in_channels=8*8,out_channels=16*8,kernel_size=3,padding=0,stride=2) #3x3\n",
        "    self.enc_cv4 = nn.Conv2d(in_channels=16*8,out_channels=32*8,kernel_size=2,padding=0,stride=2) #1x1\n",
        "    \n",
        "    self.enc_mean = nn.Linear(in_features=32*8,out_features=20)\n",
        "    self.enc_var = nn.Linear(in_features=32*8,out_features=20)\n",
        "    \n",
        "    self.dec_fc = nn.Linear(in_features=20,out_features=32*8)\n",
        "    self.dec_cv2 = nn.ConvTranspose2d(in_channels=32*8,out_channels=16*8,stride =2,kernel_size=3,padding=0) #3x3\n",
        "    self.dec_cv3 = nn.ConvTranspose2d(in_channels=16*8,out_channels=8*8,stride =2,kernel_size=3,padding=0) #7x7\n",
        "    self.dec_cv4 = nn.ConvTranspose2d(in_channels=8*8,out_channels=4*8,stride =2,kernel_size=4,padding=1) #14x14\n",
        "    self.dec_cv5 = nn.ConvTranspose2d(in_channels=4*8,out_channels=1,stride =2,kernel_size=4,padding=1) #28x28\n",
        "\n",
        "  def encoder(self,x):\n",
        "    x = F.relu(self.enc_cv1(x))\n",
        "    x = F.relu(self.enc_cv2(x))\n",
        "    x = F.relu(self.enc_cv3(x))\n",
        "    x = F.relu(self.enc_cv4(x))\n",
        "\n",
        "    x = x.view(x.size()[0],-1) #flattening it\n",
        "\n",
        "    mu = self.enc_mean(x)\n",
        "    log_var = self.enc_var(x) #for our own reference, this is nothing but a neural netwrok \n",
        "\n",
        "    std = torch.exp(0.5*log_var)\n",
        "    eps = torch.rand_like(std)\n",
        "    z = mu + (eps*std)\n",
        "    return z,mu,log_var\n",
        "\n",
        "  def decoder(self,x):\n",
        "    x = F.relu(self.dec_fc(x))\n",
        "    x = x.reshape(-1,32*8,1,1)\n",
        "    x = F.relu(self.dec_cv2(x))\n",
        "    x = F.relu(self.dec_cv3(x))\n",
        "    x = F.relu(self.dec_cv4(x))\n",
        "    x = self.dec_cv5(x)\n",
        "    return x \n",
        "\n",
        "  def forward(self,x):\n",
        "    enc_out, mu, log_var = self.encoder(x)\n",
        "    dec_out = self.decoder(enc_out)\n",
        "    return dec_out, mu, log_var"
      ],
      "metadata": {
        "id": "GXGMiLrE-gVw"
      },
      "execution_count": 26,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model_vae = VAE()"
      ],
      "metadata": {
        "id": "ynP9vkkONFQu"
      },
      "execution_count": 27,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model_vae = model_vae.to(device)"
      ],
      "metadata": {
        "id": "xnOREWhqX0aK"
      },
      "execution_count": 28,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "opt_vae =Adam(model_vae.parameters())"
      ],
      "metadata": {
        "id": "o8HrNfQyX5YA"
      },
      "execution_count": 29,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "max_epoch = 100\n",
        "\n",
        "for epoch in range(max_epoch):\n",
        "  training_loss_vae = 0.0\n",
        "\n",
        "  for data_vae, _ in train_loader_vae:\n",
        "    data_vae = data_vae.to(device)\n",
        "\n",
        "    opt_vae.zero_grad()\n",
        "\n",
        "    output, mu, log_var = model_vae(data_vae)\n",
        "\n",
        "    loss_vae = F.mse_loss(output, data_vae,reduction = \"mean\") - \\\n",
        "                0.5*torch.mean(1+log_var-mu.pow(2)-log_var.exp())*(1/(epoch+1))\n",
        "    loss_vae.backward()\n",
        "    opt_vae.step()\n",
        "\n",
        "    training_loss_vae += loss_vae.item()\n",
        "\n",
        "  if (epoch+1) % 20 ==0 :\n",
        "    print(\"epoch: {} training loss : {}\".format(epoch+1,training_loss_vae))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "rbfSZ8VVX-M0",
        "outputId": "4b8f800e-affe-41aa-e4b7-36c265930d2a"
      },
      "execution_count": 30,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch: 20 training loss : 3.0477144345641136\n",
            "epoch: 40 training loss : 2.281730178743601\n",
            "epoch: 60 training loss : 1.9348421059548855\n",
            "epoch: 80 training loss : 1.7193778567016125\n",
            "epoch: 100 training loss : 1.5755185596644878\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import numpy as np\n"
      ],
      "metadata": {
        "id": "-IH81EEDaGbk"
      },
      "execution_count": 31,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "with torch.no_grad():\n",
        "  noise = np.random.normal(0,1,size=20)\n",
        "  noise = torch.from_numpy(noise).float().to(device)\n",
        "  generated_images = model_vae.decoder(noise)\n",
        "  plt.imshow(generated_images.cpu().detach().numpy().reshape(28,28))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 265
        },
        "id": "UgLW7F6ecDw4",
        "outputId": "f535bc4f-1fc4-4405-cd19-ffbd7615fd5c"
      },
      "execution_count": 32,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAD4CAYAAAAq5pAIAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAUt0lEQVR4nO3de4xc1X0H8O939uG1d/1aG28We2OMMRRDakO3bgI0IUHl1QqTtkKhEjUKitMIpESKqiAqNfzVoioPoapCNQVhKh6iCQgX0TSuS0UhxWUxxvgBMQH8Yu0F29he2/ucX//YIV1gz+8Mc2fmTjjfj7Ta3fntmXvmzv3tnZnfPefQzCAin36FvDsgIvWhZBdJhJJdJBFKdpFEKNlFEtFcz421FtpseqEj/AcqDEyNkXgj7zev2sPYA6uhWu/TnJ6z08VBjNjQlFvPlOwkrwZwN4AmAP9kZnd5fz+90IEvdKwOxm183N9gwXkhUiz6bbPytp1VpO9sanLjbvk0tl+yPq7Yc+ZtP/K4ainTPi3n/iP/yDI9Z44XTj0VjFX8TJNsAvAPAK4BsBzAjSSXV3p/IlJbWf6trwLwhpm9aWYjAB4FED5ti0iusiT7QgD7Jv2+v3Tbh5BcS7KPZN9IcSjD5kQki5p/Gm9m68ys18x6Wwtttd6ciARkSfYDAHom/b6odJuINKAsyf4igGUkl5BsBfA1ABuq0y0RqbaKS29mNkbyNgD/jonS2/1mtsNvFCmvZSkDZS0hZSnd1bjsZyMj/h84j53NGS+lKEQKxq0tftx5vm1o2G9bw9JcrUd7Ziojx47lCo+3TEeCmT0N4Oks9yEi9aHLZUUSoWQXSYSSXSQRSnaRRCjZRRKhZBdJRF3HswOo3VDRWg/lzCKy7ehwyMjdu8M1p0cuUR72a91si7Rv8evsxfePhWMjo27bQrt/39H95tS6Mw1BBWo/pLoGdGYXSYSSXSQRSnaRRCjZRRKhZBdJhJJdJBH1L7158iyfZbnvrDORts9w48Uz57nxE0vC03MfvsAfJjrUPebG53Qfd+OFgv+cDf3is8HY4scH3LZ474gbtkjpzsacxxZ7zmLDa2PHS2yIqxfPuu1Qs4paichvHCW7SCKU7CKJULKLJELJLpIIJbtIIpTsIolorDp7nlNJx2QY0liI1NH33nS2G+clR934hWf8Mhi7oMVfcmtxm1/LPq+t34030d8v/33mucHYv6640L/v1xa48QUv+dcIdLwcXrNkbH9kPZOCX+uOPadROSxXrTO7SCKU7CKJULKLJELJLpIIJbtIIpTsIolQsoskov519izjeLPIOFbena45sqzxeI9fL26+1K91f+vcZ9340bH2YGyo6E/HHKuTHxyb7beH3/5z7fuDsYsv2uO2HV/p79dXr+tx4z/p6w3Gzru3022Ll3f58ZhaHssVypTsJN8GcALAOIAxMwvvXRHJVTXO7F82s/eqcD8iUkN6zy6SiKzJbgB+TvIlkmun+gOSa0n2kewbMf86bRGpnawv4y8zswMkFwDYSPI1M/vQp0lmtg7AOgCY3TQ/28yMIlKxTGd2MztQ+j4A4AkAq6rRKRGpvoqTnWQ7yZkf/AzgSgDbq9UxEamuLC/juwA8UVr6thnAw2b2s2grr/7YwMsuu0v4Do24bQuD/mcVg6/Pd+NPzfltN95/YlYwNjzm13uHh/06fNs0f2728aL/nHS0hZeEjrX9ysLwOH0AuLnzF278iq/sDMa+1XST2/bcfzzfjRd2h68fAACYfyzbqDMWP3acVzi3QsXJbmZvAlhRaXsRqS+V3kQSoWQXSYSSXSQRSnaRRCjZRRJR/yGuGaZkznS/tSzbtfjlK5445cbPedCfKvroC4vdeMfJ8LDh9sjw2+I0Pz48Mzx8FgBm7fXLik2nwvu92OqXBTdccYkb33OVP0z1Lz7zX8HY6pVb3bY/+0P/+rCzH/XLpeiPLEedZZlvLdksIh4lu0gilOwiiVCyiyRCyS6SCCW7SCKU7CKJqH+d3asRZqnBZx0WmKUOH7lvO33ajXPIr1XPjNVsm52nsejXc9ns17pnRurBFum7jYf3TbPXbwBn7/GXRd4y4zw3vvf6bcHYJTN3u237fu+zbvzU5nlufMaRY24c3hDXCHrLPTshndlFEqFkF0mEkl0kEUp2kUQo2UUSoWQXSYSSXSQRjTWevZZjzrPed4ZrAOy0X4tmpN5s3jLXAIDwdM1sbfWbtvpj8W04fN8AYEN+3Js63Eb8KbiLgyfdOEf9Wvhbw+GlslfM2Ou2jWkajhwPsesbMizp7E5r7oR0ZhdJhJJdJBFKdpFEKNlFEqFkF0mEkl0kEUp2kUQ01nj2WspzXnlv/DHKqKNH+sZpTi09Us+1sci4akb2S2zOfK+OH6lFD/3+cjc+93fedeOLWo8EY30nl7htB7Z0ufFz9vvbji7ZnGXe+ApFj3CS95McILl90m2dJDeS3F36Pre23RSRrMo5nT0A4OqP3HY7gE1mtgzAptLvItLAosluZs8C+OjrodUA1pd+Xg/g+ir3S0SqrNL37F1m1l/6+SCA4BsckmsBrAWANvrrholI7WT+VMomPmkIftpgZuvMrNfMelvZlnVzIlKhSpP9EMluACh9j0x/KiJ5qzTZNwBYU/p5DYAnq9MdEamV6Ht2ko8AuBzAfJL7AXwfwF0AHiN5C4A9AG6oZSfLknXd91qtG18GtkSehsh4d3c8fKQtIjV+Gx+tfNuR7R+/zK91n77ZX7f+e+dsdOOHRucEYw8/76/9fu6/HHfjOOz3DSP+fstjXodospvZjYHQFVXui4jUkC6XFUmEkl0kEUp2kUQo2UUSoWQXSUT9h7h68hyGmkVsCGpkuuboENjIVNSYEV7amDMi2/ZncwbGIsNvF/gDHvetDk/nfP51r7tt/6xrsxs/WZzmxu9+5cvB2DkPRab33vUrN442f9vxocPOc16jPGjQ7BGRalOyiyRCyS6SCCW7SCKU7CKJULKLJELJLpKIT8+SzbG2kdplbJipjYfbx5YtLnT403EVuzrd+KlFHf79j4WnJW4+7dd7W/eGp1sGAMwM1/AB4P0V89z44mvfCsbWfOZ5t+1Q0V9u+oH9/jDV7p+Ea+FNr2wPxgBEjyfveAAQvXYij2tGdGYXSYSSXSQRSnaRRCjZRRKhZBdJhJJdJBFKdpFENNaSzbFauLP8cK2XwPWWRR7+wnlu2/5L/HrxcGdkOuc2f7+0DoSfxjNe9v+ft7zjL+k81DPbjR+82p8y+c8XbAvG2ukPpv/7/f4ExsfXL3Lj8/5zZzgYWco6qhCZgyA2Jj2HqaR1ZhdJhJJdJBFKdpFEKNlFEqFkF0mEkl0kEUp2kUT8Ro1nd2vpkaWHOb3NjaNrvhs+cGU4Pvuafrft1xeGa80AMFz053bvO7rYjb9SDMcHz/Sf4uZT/nj0d77ot/+jC/vc+JymU8HYD/Zd5W/7Kf9x9zz3jhv3rryw2PESW0a76F/X4V0TAmS8LqTC5cWjZ3aS95McILl90m13kjxAcmvp69qKti4idVPOy/gHAFw9xe0/NrOVpa+nq9stEam2aLKb2bMAInMXiUijy/IB3W0kt5Ve5gcX/CK5lmQfyb4Ri6xZJiI1U2my3wNgKYCVAPoB/DD0h2a2zsx6zay3lZEPyUSkZipKdjM7ZGbjZlYEcC+AVdXtlohUW0XJTrJ70q9fBRCZl1dE8hats5N8BMDlAOaT3A/g+wAuJ7kSE6XMtwF8s+wtOrV0xtYp92qjGdc456g/v/qpVeF68V8v2eS2HYfft80nlrrxHQe63fis18JPo0WGbR9a5df4F1x00I2vaN/nxp86vCIYO/jQWW7bM9705+NHbN16pxbOVn+OARv2t81mP3XY7s+3D2etARvxx/mjwhp9NNnN7MYpbr6voq2JSG50uaxIIpTsIolQsoskQskukgglu0giGmqIa7SgkKFsF2POVNEAcFbX4WCsveCXad4f98sw295f6MbHj4SXHgaAkTlO2/NOum3/9LdeduNXzvIvoWijP5X0w6fC11t1HPDLnW07D7jx4qD/2NgUPl5szN82z+xy46MLnZ0OgKP+MNSWg8fCwSNH3bYWKROH6Mwukgglu0gilOwiiVCyiyRCyS6SCCW7SCKU7CKJaKwlmyOy1NKjUwc7Qw4BYHAkXIdvoV/39KZTBoA/6d7ixvd27nHjo8441mXTD7ltV7W95cZPmj8Edu9Ypxv/es/zwdiuvznTbbvtmH/9wc5d57tx98KNyKHUs+RdNz692Y+/9WKPG1/4THi/ztjhX7tgkTp8iM7sIolQsoskQskukgglu0gilOwiiVCyiyRCyS6SiIYazx5jWaahjo1fHgiPVweAw69cEIy9v7Tdbbu0xa/JLmv3t710Tocb/9/hcF1236i/JPPB8VlufCQ2F3XEvKbBYOyLHa+5bW/t/B83vnPRbDd+shieB+DiaQNu29bI8XTPkd9142/Br7O3HnWmkj7pX5dRKZ3ZRRKhZBdJhJJdJBFKdpFEKNlFEqFkF0mEkl0kEfWvs2fhzTkfGScfXaI3Uodftq4/GPvLmVMtdPv/brrsOTf+jbmb3fhg0V+a+EQxXOefWTjtti1G/t8vaDrhxncM+2POD42G51cv0L/m4vVhf7z77CZ/3vi2Qvj6g+dO+3Xwl08tduOPvejX2c96xh+T3vzu8WCsGDkWKxU9s5PsIfkMyZ0kd5D8dun2TpIbSe4ufZ9bkx6KSFWU8zJ+DMB3zWw5gM8DuJXkcgC3A9hkZssAbCr9LiINKprsZtZvZltKP58AsAvAQgCrAawv/dl6ANfXqpMikt0nes9O8iwAFwHYDKDLzD54I3sQwJSLY5FcC2AtALTRv4ZcRGqn7E/jSXYA+CmA75jZhz5dMDNDYHo/M1tnZr1m1tvKtkydFZHKlZXsJFswkegPmdnjpZsPkewuxbsB+MOIRCRX0ZfxnBg7eh+AXWb2o0mhDQDWALir9P3JmvSwXmJDZN87Eoyd/7d+meXJP/6SG/+3q5a78c/NC5f9AGDR9PDUwoNj/nLPp4t+SXLbYb/89W6fv7Rx2+Hwfp12xF+ku3NbuDwFAKNz/VeK49PD57LCsF/2mzbgl/WWHzvoxovv+sOWi03O0OHYMHCvraOc9+yXArgJwKskt5ZuuwMTSf4YyVsA7AFwQ0U9EJG6iCa7mT2H8JT6V1S3OyJSK7pcViQRSnaRRCjZRRKhZBdJhJJdJBGcuPitPmY3zbfPd1wX/oNYfTHDcs9RkW3Tq20WIuv/Fv19zPn+ssc2zV82eWzujGCsZY8/jbXN8i9h5ml/KWsMj/j3P+pcgxA59oqDfq27MMufBttOhIfnss2//sAiS3hz+nS//Yi/X1wZjvMXBjfg2Ph7Ux6QOrOLJELJLpIIJbtIIpTsIolQsoskQskukgglu0giGmsq6VrW0Wu4bRuPXB8QqeHbwci8H5H2TXvD1wBYZAptnIpMNR1b6joyD4A39jo2fXdhWqwW7k+x7U4fHrv2Ibbfxsf9eEwOx7rO7CKJULKLJELJLpIIJbtIIpTsIolQsoskQskukoj619ljY9Y9Xm0y61j4WC28httmkx/PVMe3DPu7DBarNztxd44AID5PwFBkLH2LPw9AJrHHXeHc7uVwjxfnuged2UUSoWQXSYSSXSQRSnaRRCjZRRKhZBdJhJJdJBHlrM/eA+BBAF0ADMA6M7ub5J0AvgHgg4nJ7zCzp2vVUQDZavS1FKujx9Z+zzge3m/rj9uOjSmPio1nd+aGj22bzZHDM1JH9+rRNhp53FnXU8iwDkF0LQf3OQ3HyrmoZgzAd81sC8mZAF4iubEU+7GZ/aCM+xCRnJWzPns/gP7SzydI7gKwsNYdE5Hq+kTv2UmeBeAiAJtLN91GchvJ+0nODbRZS7KPZN+I+dMIiUjtlJ3sJDsA/BTAd8zsOIB7ACwFsBITZ/4fTtXOzNaZWa+Z9bayrQpdFpFKlJXsJFswkegPmdnjAGBmh8xs3MyKAO4FsKp23RSRrKLJzomPku8DsMvMfjTp9u5Jf/ZVANur3z0RqZZyPo2/FMBNAF4lubV02x0AbiS5EhOf9b8N4JtlbTHP6aJrJcvw2DLax2Qq40TvPFJaq2H7zH33SlQ1fk5i9+8+tizHk3O35Xwa/xyAqZ6x2tbURaSqPoWnWRGZipJdJBFKdpFEKNlFEqFkF0mEkl0kEY21ZHOWKZkbdfhrOTJee5CpHl3DKY8zy/qcZnlsWa8HydL36BDXyu5bZ3aRRCjZRRKhZBdJhJJdJBFKdpFEKNlFEqFkF0kEM48Z/iQbI98FsGfSTfMBvFe3Dnwyjdq3Ru0XoL5Vqpp9W2xmZ0wVqGuyf2zjZJ+Z9ebWAUej9q1R+wWob5WqV9/0Ml4kEUp2kUTknezrct6+p1H71qj9AtS3StWlb7m+ZxeR+sn7zC4idaJkF0lELslO8mqSr5N8g+TtefQhhOTbJF8luZVkX859uZ/kAMntk27rJLmR5O7S9ynX2Mupb3eSPFDad1tJXptT33pIPkNyJ8kdJL9duj3Xfef0qy77re7v2Uk2AfglgD8AsB/AiwBuNLOdde1IAMm3AfSaWe4XYJD8IoBBAA+a2YWl2/4OwBEzu6v0j3KumX2vQfp2J4DBvJfxLq1W1D15mXEA1wO4GTnuO6dfN6AO+y2PM/sqAG+Y2ZtmNgLgUQCrc+hHwzOzZwEc+cjNqwGsL/28HhMHS90F+tYQzKzfzLaUfj4B4INlxnPdd06/6iKPZF8IYN+k3/ejsdZ7NwA/J/kSybV5d2YKXWbWX/r5IICuPDszhegy3vX0kWXGG2bfVbL8eVb6gO7jLjOziwFcA+DW0svVhmQT78EaqXZa1jLe9TLFMuO/lue+q3T586zySPYDAHom/b6odFtDMLMDpe8DAJ5A4y1FfeiDFXRL3wdy7s+vNdIy3lMtM44G2Hd5Ln+eR7K/CGAZySUkWwF8DcCGHPrxMSTbSx+cgGQ7gCvReEtRbwCwpvTzGgBP5tiXD2mUZbxDy4wj532X+/LnZlb3LwDXYuIT+V8B+Ks8+hDo19kAXil97ci7bwAewcTLulFMfLZxC4B5ADYB2A3gPwB0NlDf/hnAqwC2YSKxunPq22WYeIm+DcDW0te1ee87p1912W+6XFYkEfqATiQRSnaRRCjZRRKhZBdJhJJdJBFKdpFEKNlFEvF/q3CW+D+g3y0AAAAASUVORK5CYII=\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [],
      "metadata": {
        "id": "LpZdOjo6fKTC"
      },
      "execution_count": 32,
      "outputs": []
    }
  ]
}