Keystone_project.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Keystone_project.ipynb",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/souravs17031999/Federatedencryption-showcase/blob/master/Keystone_project.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mRn-DQ_UI6ZD",
        "colab_type": "text"
      },
      "source": [
        "# PROJECT OBJECTIVE : To implement federated learning and encryption techniques on MNIST dataset using Pytorch and Pysyft"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fI4HeSS_-yYC",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# let's start off by installing privacy preserving deep learning framework pysyft\n",
        "!pip install syft"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bYW5BtsKLWtr",
        "colab_type": "text"
      },
      "source": [
        "## Importing packages"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DuZtnC1v-5qw",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# import all required packages\n",
        "import torch\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F\n",
        "import torch.optim as optim\n",
        "from torchvision import datasets, transforms"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "W6j1H-R-LaUi",
        "colab_type": "text"
      },
      "source": [
        "## Creating virtual workers"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1N5I5MWi_AHK",
        "colab_type": "code",
        "outputId": "31e3375f-9b9e-45dc-dfe0-5a1850be1c57",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 107
        }
      },
      "source": [
        "# creating virtual workers\n",
        "import syft as sy\n",
        "hook = sy.TorchHook(torch)\n",
        "# let's create two virtual workers who will hold the data while training the model locally\n",
        "bob = sy.VirtualWorker(hook, id = \"bob\")\n",
        "alice = sy.VirtualWorker(hook, id = \"alice\")\n",
        "crypto_provider = sy.VirtualWorker(hook, id=\"crypto_provider\") "
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING: Logging before flag parsing goes to stderr.\n",
            "W0813 02:05:57.266631 139640065849216 secure_random.py:26] Falling back to insecure randomness since the required custom op could not be found for the installed version of TensorFlow. Fix this by compiling custom ops. Missing file was '/usr/local/lib/python3.6/dist-packages/tf_encrypted/operations/secure_random/secure_random_module_tf_1.14.0.so'\n",
            "W0813 02:05:57.284585 139640065849216 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/tf_encrypted/session.py:26: The name tf.Session is deprecated. Please use tf.compat.v1.Session instead.\n",
            "\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "aiivamvOLeeH",
        "colab_type": "text"
      },
      "source": [
        "## Setting hyperparameters and device"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0N4vXtvV_M6R",
        "colab_type": "code",
        "outputId": "6b2f6799-8317-49ad-8189-a5d2fe5fdf62",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "# This arguments class is simply defining all the hyperparameters of the model we are going to train on\n",
        "class Arguments():\n",
        "    def __init__(self):\n",
        "        self.batch_size = 64 # batch size for training\n",
        "        self.test_batch_size = 64 # bacth size for testing \n",
        "        self.epochs = 10 # no of epochs\n",
        "        self.lr = 0.01  # setting learning rate\n",
        "        self.momentum = 0.5 \n",
        "        self.no_cuda = False  \n",
        "        self.seed = 1\n",
        "        self.log_interval = 10\n",
        "        self.save_model = False\n",
        "\n",
        "args = Arguments()\n",
        "\n",
        "use_cuda = not args.no_cuda and torch.cuda.is_available()\n",
        "\n",
        "torch.manual_seed(args.seed) # sets the random seed from pytorch random number generators\n",
        "\n",
        "device = torch.device(\"cuda\" if use_cuda else \"cpu\") # choose device based on what is available if cuda or not\n",
        "\n",
        "print(device)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "cpu\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6iKEU6JGLizm",
        "colab_type": "text"
      },
      "source": [
        "## Creating federated dataloader for loading of data and distributing it across the workers"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_PkLH7Il_NTL",
        "colab_type": "code",
        "outputId": "6b2932d5-b5fe-49b1-85f9-4cc5c5dc1d55",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 263
        }
      },
      "source": [
        "# using .federate method, we can distribute our data across all the workers for now , we have bob and alice\n",
        "federated_train_loader = sy.FederatedDataLoader( \n",
        "    datasets.MNIST('../data', train=True, download=True,\n",
        "                   transform=transforms.Compose([\n",
        "                       transforms.ToTensor(),\n",
        "                       transforms.Normalize((0.1307,), (0.3081,))\n",
        "                   ]))\n",
        "    .federate((bob, alice)), \n",
        "    batch_size=64, shuffle=True)\n",
        "\n",
        "validloader = torch.utils.data.DataLoader(\n",
        "    datasets.MNIST('../data', train=False, transform=transforms.Compose([\n",
        "                       transforms.ToTensor(),\n",
        "                       transforms.Normalize((0.1307,), (0.3081,))\n",
        "                   ])),\n",
        "    batch_size=64, shuffle=True)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\r0it [00:00, ?it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Downloading http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz to ../data/MNIST/raw/train-images-idx3-ubyte.gz\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "9920512it [00:01, 9575845.36it/s]                            \n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Extracting ../data/MNIST/raw/train-images-idx3-ubyte.gz\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "  0%|          | 0/28881 [00:00<?, ?it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Downloading http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz to ../data/MNIST/raw/train-labels-idx1-ubyte.gz\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "32768it [00:00, 135435.06it/s]           \n",
            "  0%|          | 0/1648877 [00:00<?, ?it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Extracting ../data/MNIST/raw/train-labels-idx1-ubyte.gz\n",
            "Downloading http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz to ../data/MNIST/raw/t10k-images-idx3-ubyte.gz\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "1654784it [00:00, 2229496.99it/s]                            \n",
            "0it [00:00, ?it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Extracting ../data/MNIST/raw/t10k-images-idx3-ubyte.gz\n",
            "Downloading http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz to ../data/MNIST/raw/t10k-labels-idx1-ubyte.gz\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "8192it [00:00, 51927.48it/s]            \n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Extracting ../data/MNIST/raw/t10k-labels-idx1-ubyte.gz\n",
            "Processing...\n",
            "Done!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "B6rIap2tMMdB",
        "colab_type": "text"
      },
      "source": [
        "## Visualizing dataset"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2foszLff_PtV",
        "colab_type": "code",
        "outputId": "ecc8548c-61d5-4759-c82b-6c231ff9154c",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 383
        }
      },
      "source": [
        "images, labels = next(iter(federated_train_loader))\n",
        "\n",
        "# Checking shape of image\n",
        "print(f\"Image shape : {images.shape}\")\n",
        "print(f\"Label shape : {labels.shape}\")\n",
        "\n",
        "\n",
        "# denormalizing images\n",
        "def imshow(inp, title=None):\n",
        "    \"\"\"Imshow for Tensor.\"\"\"\n",
        "    inp = inp.numpy().transpose((1, 2, 0))\n",
        "    mean = np.array([0.485, 0.456, 0.406])\n",
        "    std = np.array([0.229, 0.224, 0.225])\n",
        "    inp = std * inp + mean\n",
        "    inp = np.clip(inp, 0, 1)\n",
        "    plt.imshow(inp)\n",
        "    if title is not None:\n",
        "        plt.title(title)\n",
        "    plt.pause(0.001)\n",
        "\n",
        "# plotting the images of loaded batch with given fig size and frame data    \n",
        "import torchvision\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "grid = torchvision.utils.make_grid(images.get(), nrow = 20, padding = 2)\n",
        "plt.figure(figsize = (20, 20))  \n",
        "plt.imshow(np.transpose(grid, (1, 2, 0)))   \n",
        "print('labels:', labels.get()) "
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "W0813 02:06:30.004012 139640065849216 image.py:648] Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Image shape : torch.Size([64, 1, 28, 28])\n",
            "Label shape : torch.Size([64])\n",
            "labels: tensor([6, 8, 1, 7, 3, 0, 7, 1, 1, 2, 2, 6, 0, 8, 1, 5, 0, 1, 1, 4, 8, 1, 0, 9,\n",
            "        3, 5, 6, 7, 0, 1, 9, 0, 7, 6, 0, 3, 2, 7, 7, 3, 8, 6, 9, 2, 2, 0, 9, 9,\n",
            "        7, 0, 4, 3, 2, 9, 2, 7, 2, 7, 1, 4, 8, 4, 7, 9])\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAABIsAAAEFCAYAAABjDWdhAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4zLCBo\ndHRwOi8vbWF0cGxvdGxpYi5vcmcvnQurowAAIABJREFUeJzs3XvgFXP+x/HnKC1CaeWSsqJEorSt\nJHcRtlx+7iKXkqXWtVbIXaRVYos2FWUT5bJksa7lniRESQlbiFxiQ6id3x/1Od/zPd+Zc+bMfc73\n9fiHZuZ8znvOZ86c+X7mPe+PZds2IiIiIiIiIiIiAOslHYCIiIiIiIiIiKSHBotERERERERERCRH\ng0UiIiIiIiIiIpKjwSIREREREREREcnRYJGIiIiIiIiIiORosEhERERERERERHI0WCQiIiIiIiIi\nIjmRDRZZlnWoZVkLLMtaZFnWwKjeR0REREREREREwmPZth1+o5ZVB/gAOBhYCswCTrJte17obyYi\nIiIiIiIiIqGJKrNoD2CRbduLbdv+BbgPODKi9xIRERERERERkZDUjajdbYAlef9eCnR029iyrPDT\nm0REREREREREarevbNtuXO6LohosKsmyrD5An6TeX0RERERERESkwn3i50VRDRZ9CjTL+3fTdcty\nbNseA4wBZRaJiIiIiIiIiKRFVDWLZgEtLctqbllWPeBE4NGI3ktEREREREREREISSWaRbdurLcvq\nB/wbqAOMt237vSjeS0REREREREREwmPZdvJPgOkxNBERERERERGR0M22bbtDuS9KrMB12BYuXAhA\nixYtsCwr4WhEREREpFxuNzF1bSciIhKvqGoWiYiIiIiIiIhIBlVMZlGLFi1y/3/FFVcAcN111yUV\njoiIiERs6NChAAwYMKDodg0bNuS7776LIyTxafXq1UmHICIiInmUWSQiIiIiIiIiIjkVk1kkIunT\nq1cvxo4dW/brevfuDcC4cePCDkkkkPbt2wMwe/bsasuzUE8lvxZMFuLt1q0bAD169ADgT3/6EwAr\nVqwou60VK1bQsGFDgFRmGJnj6qyzzgLgnHPOSTKcRNSpU8dxeRaOVT/efvttdtttN8d1S5YsAWDb\nbbeNM6RA1l9/fX755Zei23z22WcAbLPNNnGEFDrV05Kkmd/FadOmVVteG47BBg0aADWvAdKw74Xn\nhjTEFBZlFomIiIiIiIiISE7mM4vatm1bY9ncuXMTiKRyZOHOyYgRIwA4//zzc8v69+8PwLBhwxKJ\nSWryk1WUVW53PCBd3x0JpjCjKAvOPffcpEMoi9tv0IknnhioXfPdTPr7eMghhwDw73//23Ubk0Vl\n3HfffQCcdNJJ0QWWkAsuuCDpEGLhdlw7adasWYSRRKNUVhFAkyZNgLWfRdLfQwnPrFmzAOjQofqs\n3Fno4zZt2gBr/3bcddddc/8PVb8Zm222WTLBOSjMKDLM38Nvv/12nOHEyk9WcdTMsV/InO+z8B0o\nRZlFIiIiIiIiIiKSY5VzpyOyICzLdxBO8VfCKF6SspBZ9NJLLwHQuXPnGuvSFGe5JkyYAEDPnj1r\nrKtXrx4Av/76a6wxBeF0LJl6RF6yjrLQl+WcQ7OwP+LMrZ+HDx8OwMUXXxxnOGUxmUWjRo3KLUvj\nsTh58mTAXwbRpEmTqv3b1DlyktS+e8ko8iqN/eeX23frmmuuAeDqq6+OMZrwHHbYYQA8/vjjgdtK\nY383btwYgC+//NJ1mzvvvBOAPn36ANm7ZjeZfPfee2/R7bbbbjs++eSTOEJKDS/XPmnu26xdu3n9\nfUxDrGFzuoYB6NKlCwDPPvts7DEZpY6jlPXHbNu2O5TerDplFomIiIiIiIiISI4yiypcqf7t3r07\njz32WLVlWai0X2y/0hSnm9deew2AL774AoAjjjjC82tHjx4NVN5sOVn7Lgc5d6Z5v6Sml156yTGL\nEbLRl1nJLCpW96tQqfiD/EacfvrpANx9990l4yhHsZgWLFgAQKtWrTy1dcIJJzBlypRQ4kpS1n/L\nnbz33nsAtG7dOrQ2TbbxPffcE1qbQQX5zU77zIzFsrydbLjhhqxatSrKkBLn5ZrHnMdMxu2YMWMi\njcmLnXfeGYB58+b5biMNx2htzixK8xMvaY7NgTKLREREREREREQkGGUWVRhzp/G4447z/JrCzysL\no6RZuxtpZnu55ZZbQmszjfvpR9b6MsxzZtj7Vxhb1J9f1vquXMXqVWRp/7KSWRSGNNfReOKJJwA4\n9NBDPcdx6qmnAjBx4sQa67Lch8Xu9pt9Pe2002KNKaj69esDsHLlyqLb5fdb06ZNAViyZAkA7dq1\nA+Ctt94q+dqkLF++HIDNN9+82vJyYktzZtEhhxxSdl2xtO2DX7169WLAgAGA9yzHww8/PHduS4sN\nNtiAn376KbT20tC/yiyqKQ37mubYHPjKLKobRSRxcOucRYsWxRxJenTr1q3GIFH37t0Bajxq5vT5\nZeGAP/7445MOwRc/g0Tmcx85ciQAffv2rbbeTI9ppsvMmjQMVJfj+uuvD73NqKfWDHvwyDw+2bFj\nR8/vnabzR7mcBon++te/JhCJhGHw4MGJvr8peFwO89iR02DRZZddBsANN9wQLLAEFHskJGuDRLB2\nYLlUEWSnc+HSpUtd16VV4SBR0t+rJGWp36Bq8NlwOq+UcuyxxwLw4IMPhhJTGBYvXgxA8+bNE44k\nOnXq1Ek6hESkOSnkueeeq7Fs5syZob7HrFmzAOjQofoYT5yfgR5DExERERERERGRnMxmFrkJOqL3\nww8/ALDRRhs5rr/mmmtSO41rfkHqqVOnAjUzipxkIaPIuP/++13XZfnu1tFHHw3Ak08+CVCtUGJh\nRlFWZS2TqNDll1/uuPzCCy8EnLPHNt54Y6DqvOKWmRNXFo55H1NYHWCrrbYq+pply5ax5ZZbRhpX\n2jgdq7feeisAf/nLX+IOJ7DC6WZrq0GDBiUdgm9mGvmrrroq4UiCeemll1zXpfGao5T1118fKD61\nejn75XY+vu+++8oLLEblfK+ycB3w1FNPMWfOHADat28PZCPuYoLEX/hZpEGbNm0AmDt3buC2zHVa\nqcdHk7ZmzZqkQ4iVuV7O98477yQQibsDDjigxrLVq1eH+h6FGUVJUGaRiIiIiIiIiIjkVFxmkR+X\nXHIJQ4YM8bTtVVddlbuzl5a7YGaq+3xBavukZb/Klea7xuYzNVPqfvbZZwAcfPDBrq/53e9+V7TN\nLNQq6tWrF2PHji3rNb1792bcuHERReSPmdLbyYgRI6r9t5g999wTqJpO9qyzzqq2fr/99mPGjBl+\nw/QsP1Mo63dMw/TQQw+5rjNF6iWdSh3HWf1dqxSNGjUCoHPnzjXW7brrrnGHE5pffvnFdV3r1q3L\nbu/zzz93XG6K7WdVmuuOOCmVRTN+/PiYIvFv1qxZvrIS0twvhxxyCEDZBcjzpXn/gvj555+TDiEU\n5lrLqTZm2v7uWbNmTY1aUk6/cVmnzCIREREREREREcmp1ZlFQe+om7vQ//d//xdGOL717NnT87ZZ\nneraTC+bdbvssovnbT/++OPoAkmhNB9/K1asqLEsSLx9+vQBqp53btGiBQDTp08P1K55rbKFyqOs\nlOyqLce6U62iLM2C9vXXX7uue/fdd2OMJBymZooTU99u/vz5cYWTqKOOOgqAf/7zn7llp59+OgB3\n3XWX42uyck4dOnSo4/JevXrFHEn5nLKKfvzxRwDq168fdziB+T3XZ+VYC2qDDTZIOoRQONX/NHVd\n06Zu3bqRXoOk5fpGmUUiIiIiIiIiIpJTKzOLwhqpS+tIZ74pU6YAcNxxxzmunzp1aqD6RnFZsmRJ\n0fXnnXdeTJEkb+DAgUmHEKrevXsnHUJijj32WADeeuutUNs1d9JMrSWnzCgpXYeoWbNmMUUi5fL6\nO571u8pu+xl05te4mFpFbrLaP06zMAXZlyzNSlvo4YcfLrnNV199BUDjxo2jDidUAwYMSDqEUHTt\n2hVYO9NbVpi6nX6y7OvVq8evv/4ackSSlPysxdogLRlFhjKLREREREREREQkJ7OZRSarp/CORo8e\nPTjllFMcX+NnNonCuzqLFy+mefPm1ZaZEcCk7gA5ZQ3Vthocf/vb35IOIVTF+u+mm26KMZJg8mc1\nc5sVzSxP2wxoUHwWtDD06NEj0va/++47oOr7/vTTTwPQpUuXstq57777ALjyyisBWLhwYbX1absL\nUsp66629T+L0bDyQmx1z6dKlscUUhauvvjrpECLhVkckX9Z/41577bWi67OSTVusVlEWuZ3rtthi\nC1/tZbHOyPDhwwG46KKLSm6b9e9hpTCzh2WpP/xkFG288cYAZWUVFbt+efLJJ8uOISqFs25VGrd+\n2H///eMNJAFXXHEFANdee23JbZP4DiuzSEREREREREREcjKbWeTn+cV777235DaFI3bt27cHYPbs\n2a6v6d+/f9mxhKl79+4ATJs2zXWbqVOnAmSiPpE4GzVqVNIh+GIyhsx/s5SF4jSbSJij+oX1EKK+\ng3LwwQfXWGbqAjRp0qTa8kWLFgGwfPnySGOKW9u2bV1rRJmMoksvvTTOkCJz4okn1lhmZmrKsmJ1\nRLJ05zxf27ZtAXjggQeAqhkSC5lzxOuvvx5LXFEpp5/Mb0br1q2BdM4y5vc82apVK8fl//rXv4KE\nEwkvv91Z/f5Vop49ezJx4sRqy0wfmr9bhg0bFntcpfi5RqxXrx4QXkaRcdhhh5UdS1TWrFlTdH3S\nT7lEZcaMGUmHUJTpl3Iyv/wc4+XMfB62zA4WFeP2hZkzZw4Au+++e43XBCkIm/TJttggUaWdNCrd\nhAkTXNf169cvxkiiYwpaFz6WZr63vXv3Ts0jabfddlsk7aZpwOyTTz6p9t9K5zRQtNdeewHw6quv\nxh1OpJz+EM3SdOuF0vS9CcPixYsBajza7mT8+PFA+i+c/XjzzTcB52uzfPPmzcv9f9zXNlEdex07\ndnRc3q1bt0jez49K+955ceqppzoub9euXcyR+HfPPffkHo+8+eabq60z/zb/zerfCn7i9nI8m2sC\nid6IESMcl2flmKxbd+1QSuFxVUnnTT2GJiIiIiIiIiIiOZbfkS/LspoBE4EtARsYY9v2rZZlNQLu\nB7YDPgaOt2372xJtBR5+K7Yf11xzDQBXXXVV0LepxhT5u/jii0NttxRzx6k2ZRS59W+lZQQUO45r\nS5+mKbPIKcYw+iHL0yTny9JjCU6xDh48GIBBgwbFHU4sojp+4zZy5EgA+vbtW2PdpEmTAFwntkij\nIHccTaZAVh4PLXWuC3AN6jsmP6I6Zxe2ax7/bdmyZaB2w+B1n/O3y+L5xUml/EYXysLkN+WcE7zE\nW057JnPs7bff9vyauEyePBlwfrw8Xxr6sByV8l2LOpMopM9jtm3bNetrlBAks2g1cLFt262BPYG+\nlmW1BgYCz9q23RJ4dt2/RUREREREREQkA3zXLLJt+3Pg83X//1/LsuYD2wBHAvuv22wCMB24JFCU\nAYWdUWSkMaOo0piRdDeVklFkigw7KXea87SrpOd4y1Upd1Cy5LLLLnNdV6kZRVtttVXSIYTiueee\nA+CAAw5w3aZHjx7V/luMqQ+TVHHoMM59X375ZY1lWTx/ePksTBaV0z5XAqfPIA0ZReA/K/H6668H\nKvfcmnXbbbcd4D4tfdOmTVm6dGl8AQUU5vVkx44dU5lRJNlgzo/HHHMMUDVZRVDHHntsKO0EEUrN\nIsuytgN2B2YCW64bSAJYxtrH1EREREREREREJAN81yzKNWBZGwMzgMG2bT9kWdYK27Yb5q3/1rbt\nzRxe1wfos+6fvw8URJ6osxY23HBDVq1aFel7FPL6jHFtem7cqA37mcV9DPI9rOSaRZWWWZSFmkWV\nUrenHCaz6PPPP6+xLq373qBBg7JmIQ3DlVdeCcB1110Xy/sV+774mX7XjVMtLlPjaPPNN6+2bdTH\ng59zhMmuMTM1HXHEEdXW77jjjixcuDCkCL0J89y9xx57ADBz5szcsh133BEg9v0q9MQTTwBw6KGH\n5paV2sf33nsPgNatW3t+TRodf/zxAIwaNarG98RN1vYzS9fUcWWhZ632aamnTJ588kkADjvssNhi\n8sOtf3/88UcA6tevH2c4kQlhnCWkSIAEahZhWdb6wIPAJNu2H1q3+AvLsrZet35rwDF/2LbtMbZt\nd/ATtIiIiIiIiIiIRCPIbGgWa2sSfWPb9gV5y/8KfG3b9hDLsgYCjWzb/kuJtkIdPt57770BePHF\nFwO3leQou1vfdO/eHYDHHnsMcB5lTtPdgSCydBckCKf9NHfp5s+fH3c4JUV5xydNfRo0K8XcIf/g\ngw8c16dpX/1QZlG6Fe772WefzZgxYxKKprg01DKL67g46KCDAHj22WdLbjtlyhQAOnfuDECTJk1C\niSGufb3ggrWXh7fccktobSbx/S3nWuTTTz913KZY36XlnJS/n0OGDAHg0ksvddzWZOLcf//9uWWF\n16dpYvolrO8QpKff3JjaqiZLz029evUA+PXXXyOPySuTZdeiRYtQ2506dSpQdfxmjdf6tXPmzKF9\n+/ZxhORLbfn7rhyFn8nLL7+cG9MIia/MIt8FroHOwKnAXMuy3lq37DJgCDDFsqxewCdANr+NIiIi\nIiIiIiK1UJDZ0F4C3Ib9DvLbbhheeukloOaoZDl3LpMe0SznbnglZhSZWXBqszRmFEWpd+/eSYdQ\nwzXXXANUn1HR3I0yd/zdPPTQQxx99NHRBSfiwm1GpebNm8cciTjxklFkuN39njBhAj179iz7vU1t\npLiMGDECCCezKMnrmx122AGADz/80HF9GjLjgjB1h/KZjKL11ltbseKXX34B3OtqLVmyJJUZRVH0\nzaJFi0JvM2zlzDiYpowiw/yOmVmRTzzxRF/tlMqQy5p58+Z52i4Lx6hkQyizoYmIiIiIiIiISGUI\nPBtaKEGEXLOoEuT3i9sz4JVci8NkbRx33HFFt8v6/o4cORKAvn371liX5n3zc94wmUNpmemsHLNm\nzaJDh+KP+U6aNAmAHj16lGwvzX1bjqzWLCqUdIxhM7OIrFy5ssa6gQMHAnDTTTfFGpMX77//PgCt\nWrUCqu6g7rLLLoHbvvHGG3P7XqhS+r/Y70nWZvxJq7CvmdNy7DnNaOZVWvbBTa9evQAYO3as4/oF\nCxYA8PDDDwM4nifSvo8A//jHP4Di1yBZnmlq++23z2W1udWBNLMKQvIzC0btxhtvBJyPV4CuXbvy\n1FNPxRlSWSptduAwpLVmkTKLREREREREREQkR5lFKeWnXypxNLbSq+W77d+FF16Yq/WQRlmq/xWW\nIOfKn3/+GYANNtggrHBSIUtZO/mxZvnuajmuvvpqoKrmVppnQxPJApMZe+aZZ5b92o033hiAH374\nIdSYwlROnZvly5dHHU4i2rRpw9y5c4G1d/aBsO/uR8Kp79555x0A2rZtG3c4EpOsZugUxm0y/Hba\naackwkmFtGYWabAoxbz2TdpPCEE89NBDADUKBVfKPmf1JC8wdOhQAAYMGOC4fsGCBRX/o1fsHKVj\nWEREJB6Fv8f6DRaRAnoMTUREREREREREglFmkUiCHnnkEQCOOOIIAEaPHg3AOeeck1hMIl4ps0hE\nREREJPWUWSQiIiIiIiIiIsEos0hEREREREREpDIps0hERERERERERILRYJGIiIiIiIiIiORosEhE\nRERERERERHI0WCQiIiIiIiIiIjkaLBIRERERERERkRwNFomIiIiIiIiISI4Gi0REREREREREJEeD\nRSIiIiIiIiIikqPBIhERERERERERydFgkYiIiIiIiIiI5NRNOgAREcm29u3bAzB79uzcspNPPhmA\nyZMnJxKTiIiIiIj4p8wiERERERERERHJUWaRiIgEkp9RZHz22WcJRCIiImnx2GOPAfDHP/6x2nLL\nspIIR0QkVG3atAFg7ty51ZZX0jlOmUUiIiIiIiIiIpJj2baddAxYlpV8EJIpTZs2BWDJkiW5ZZMm\nTQLglFNOSSQmgffffx+AVq1auW4zfPhwAIYMGQLA8uXLow9MIlHs96OS7qpkjVu/1MY+Wbx4MQDN\nmzfPLauNn0MWPPfccwAccMAB1ZbvtddeALz66quxxyTBuJ2L7rvvPk466aSYo/Fmv/32Y+zYsQC0\naNHC8+t0XkmnN998E4Ddd9+96HbqP/Gj1DhKyo6r2bZtdyj3RcosEhERERERERGRnFqdWeRn3xs2\nbMh3330XQTTihZc+++9//wvApptuGnU4sk4Y55GUjb6XrVu3bgBMmzat7NeuWLECgAsvvDC37O67\n7w4lrihk7E5KreHWL7U5MyP/M/nqq68AaNy4cVLhSBGlzis9e/YEYOLEiTXW6ZyTLln8jVhvvbX3\nz6+99louv/xy3+2kcd8K/fnPfwbgtttuK7ntyy+/DMDee+8daUxh83NdmoW+k3Qp5zhr164dAG+/\n/XZU4ZSizCIREREREREREQmmVmYWBd3nRx99FIAjjzwyjHAiMWHCBKDqTpybrIyi++mzn3/+GYAN\nNtgg7HBCs/322wPw4Ycfltw2zX0V5nkkzftZTFTn0jR9HqX2cdSoUQD069cvjnCkQGH/tG7dGoD5\n8+cnEU6inI7VNH2XwrB69WoA6tSpU3S7tO+3fj9KO/fcc3P/f/vttycYSXFe+zKN/TRr1iwAOnQo\n+8Z7Nffddx9AqmoyjRgxAoDzzz8/cFvjx48HoFevXoHbikqx47B///4A3Hzzza7rhg0bFk1gUnFO\nP/10AO66666yX5vAeTCZzCLLsupYljXHsqzH1v27uWVZMy3LWmRZ1v2WZdUL+h4iIiIiIiIiIhKP\nwJlFlmVdBHQANrVtu5tlWVOAh2zbvs+yrNHA27Zt31GijVgyi8K+85/GOyOG275mrWZDpdXC2WOP\nPQCYOXOm7zbStD+G6aeuXbsC8NRTT7luO27cOADOPPNMx/Vp3D8vKjmzqE2bNgDMnTu36HZpiDUM\n7733HlCVmZMvzftojkFT0+W0006rsc1//vMfAJo1a1ZteQqepQ/F999/D8Amm2xSbfnJJ5/M5MmT\nkwgpVGGdZ9J0HCuzqEqDBg2Aqjp2TtK6j9dff33JWj9pjN3U4nnxxRdLbmviX7VqFQC/+c1vSm6b\npCifHknD/hVy2t9SGc/5r0lrZlGbNm1c/24wmaXFjsVCaew7Lz799FMAmjRp4vk1ce1rkO/aO++8\nA0Dbtm3DCseNr8yiukHe0bKspsAfgcHARdbaHjkQOHndJhOAq4Gig0USHrdBoN/+9rcAfPPNN3GG\n41sMX5hYXXzxxYBz2itA7969gaqBlHzffvstsLa4OlSdkNJ0si8nFpO67DZYlDW1YZrySh0kGjp0\nKAADBgzw/Jqnn34agIMPPjiSmMJgLjwK3XHHHTUGiYy33noLgI4dOwLw+uuvRxNcxAoHiYysDhSV\ncwHavXt3AD7++GMA+vbtC8Cf/vSn0ONKi6233hqAzz//POFIYKuttgJgyy23BNwHXo855hhgbXHh\ncv7oATjhhBMCRBiN3/3udwBFB4rS/BsxZ86cktsUxl9Y3sDpe2oG/tI4KY6X/jDHcxq+W6Vccskl\nNZadccYZQPomC1m2bFnuHFHbJFH6xjwmFvVxYL5Thd978zuw2267ub7WrDOfz5133glAnz59ogm2\nTEEfQxsB/AX437p//xZYYdv26nX/XgpsE/A9REREREREREQkJr4ziyzL6gZ8adv2bMuy9vfx+j5A\nLENmQaZPLPbaww47DIAnnnjCX2DiauDAgUmHECq3jCIvd3dee+01AA499NBQY5Jo/Pjjj0mHEKpi\n58ALL7wwxkjCMXLkyFzGhR9dunQJMZpwHH/88Z62y88wWbRoEQAtWrSoto1JdU9zJoATM6lDoazt\nx8iRIwFKHqM77rgjCxcuLLqNU0ZR1j4PN6bI7rJly4Dk9yvqO+Y77LADAIsXL470ffwwmWxOku4X\nL1auXBlJu2nIKPLz+Zcz8UpamBIb+dKWUWTUxqyiRx55JPb3nDRpEhD/cVD4vc9/Usbr78RZZ52V\n+28azqFBHkPrDBxhWdbhwAbApsCtQEPLsuquyy5qCnzq9GLbtscAYyD+2dBERERERERERMSZ78Ei\n27YvBS4FWJdZ1N+27R6WZU0FjgXuA04D4h9OXCdIEWcvo3/KKIrOiSee6Gm7/BFXtz67/vrrARg0\naFDwwHwIOoWzucNceJfHjFZnpSBtqeLef/3rX+MMJzL169cH1j7vf8sttwAwduxYAJ599tnE4iqX\nlztBZjreNPrhhx8A2GijjXy34ZRhmsbMsSlTpgBw//33l9x2+PDhQFUdNSOJWgJh6tmzZ9IhBHLj\njTcC7hlF5UzHHfQ3J05+j7sDDjgg5EiC6dixY6CJKwqlrWaFk2J9l9bjrVxB9uPNN98EoH379mGF\nE4tSGUVp7FtT79NMX75mzZqS9T3N3wYSvQceeACAI444wnUbL8fV+uuvD8Avv/xSdLszzjgjlZll\nhftojsFSEwMkLWjNIieXsLbY9SLW1jCqWbFXRERERERERERSKdBsaIZt29OB6ev+fzGwRxjt+rXz\nzjsDMG/evEjav/XWWyNpNwxepv6U5Pi5I5PGGgVeXXHFFVx77bWetv3LX/4ScTThcrur6rTcZMqt\nWbMGgLp1Qzn1RsKcP4PeAUpKkAwZL/uVhQw4twyjcvrNzIRjasKk1VFHHeW4PM3HaKGLL77YtU5f\nOfvhduxvscUWvuKKkpfvaWF2TZqz315//fWyjzmn/dluu+0A+OSTT8IIKxLFZhY0sylmQbHjqV27\ndp7aKPZZFKvllDaNGjUC4Ouvv3bdJgvn1P/9b+18S14yS5yyOYYNGxZFWNU0a9aMJUuWhN6ul1q7\nSbnnnnty/zWZdibzrhylMorSeIzmP+X05Zdflv16M/PiqlWrQoupXFFkFomIiIiIiIiISEZZaRiB\nDLvAddT7lMaRS8Nt39MccyEv/ffFF18AVXe/i73u5ZdfBmDvvfcOITrvzOj57NmzczMPtWzZ0nd7\nhftn7nyloWZR06ZNAcq6W5I5bKpOAAAgAElEQVSFLBsnYZxfTjjhBKCq1kyaZL0ORan+efTRR3Of\nv9c7NfltpvkzMHH2798fKO8uaeHnNnjwYCC5Wm9eVepv3oYbbgh4O0bdPoMnn3wSqJq5NQ28nD8L\n+85k1zZv3jy37KOPPgKqZm7KEjOLXeFMhJCN49atD9944w3+8Ic/xByNf277ceyxx/Lggw8GagOy\n3Zf5srAffqShtlthDZ6uXbsC8Pzzz9fYdsiQIQBcdNFFvt/vmWeeAeDggw/23UYSSh2nEydOBOC0\n006LIxxPzEzWYWVbhnRszrZtu0O5L1JmkYiIiIiIiIiI5GTrlr4HY8aMibT9Sh1hTwMvdzgWLFgA\nwE477eS53c6dO/uOKSxOdxC9CpKNFBc/z1/XqVMHqNo/c7c1rfxkFHXv3h2AadOmVVtuasqkKbOo\n2P6ZmZhKWb16NbC27sbSpUtDiatc5hxtajB88803vtt6//33Q4lJ4vPOO+8kHYJnL730Uo1lo0eP\nBkpnFDVo0IAVK1YU3SYNGUWnnnoqUHXntxi366v8jCIj7bPHOKn0LJSsZBWV+t31klV00EEHhRVO\nrMw54fHHH3fdJq0z8RU7/iZNmgTAKaecElc4ofj1118Bb99/M4Np4UymV1xxBYCn+qBZyyiC4v2+\n6667AvDuu+/GFY5nKcsoCkSZRSIiIiIiIiIiklNxmUVRScPInpSXUZQ0p0r/ZoR8//33B2DGjBmO\nr915552ZM2cOAL/5zW+iCTAlPvjggxrLsvx9c4rdbZaKb7/9FoDNNtss+sBcmBjcTJo0yfVundsd\nn/xMs6T6MkhGkdGqVasQIonfTTfdBMQzs0tS3GaJbNu2bcyRlM/M4OaU9XrOOec4vuYf//gHAD16\n9CjZfprOn0Eyijp16uT6mmKzUKWN23ny1ltv5YILLqi2zMx889NPPzm+5plnnok9O+COO+4ouj6/\n/wr3Z8SIEZHE5Me5554LwHHHHee43sv3xtSiNLVfsqKczOizzjoLSD6zyGSdO10jFjLnxfzzo5m5\ntHCmXbfzSrNmzXzFmTQvmf316tWLIZJwFbs2NZlkacwo8jPLWzGnn346QMnZ/aJUcQWuo9qfNF18\nuRk5ciR9+/attizNhXQLFeu7vfbaC4BXX33VdztJ9mEUx2WaClwbTtNYP/zww77bS9P3rlgxTPCW\nuu7WVpqPTafY/BzPaepLr/L388ILLwTS9cdPoSDH1SWXXAJUFdFMe4Hrwn295pprALj66qsTiKY8\nYf8ezJw5E4A999wz1HaD8LqPxY5RtzbWrFmTiYkRovjdT+I86rYf5lFyM12503bjx48HoFevXhFF\n5525Obfvvvs6rvdyji88Tzox6y699FJfcYYpjGMw7mPObZBo1KhR9OvXr9qyY445BoAHHngg8Pua\nR5rypXEwopCXPjaDReaxtyyo9Md2Ye0Ng3znn3++67YqcC0iIiIiIiIiIqlQMZlFjRs3BuDLL78M\nHI8Tk7Fz++23R9J+GJwyi7Iw+jp06FAABgwYUGPdySefDJSXcp7GzKJCTlMBFzIptE6fC6Qzs8iP\nUnf7IB19N2vWLAD++9//AnDggQf6bisLmUWtW7cGYP78+blHC2655Rbf75OGPvQqDVPq+hHkuCp8\nbdYyi7bYYgsAli9fnkQ4njz99NMAdOnSJZT2zj77bCD6iT3KEUZGUam2tthii9T180EHHZR7bHXU\nqFFlvz6NBXq9Zp1mZfp1P1m0Xl8LVRNBnHTSSeUFFqEGDRoA1CiGbyYCMI/tFnv8Me6+C3IN76WA\ntx9pOH4LNW3aFPD2GFoa43fz3nvvAVXXoPmytB9BRHgNqswiEREREREREREJJvOZRWFkFFmWFerd\nsKRUYmaRn/jNlMOFhaGz8FkUU3iMVkpmUaHVq1fnaiEYvXv3BmDcuHFJhBSaCRMmANCzZ09gbe0N\nINH6G17u4gX5ncji906ZRene36z2j+Hl+3TeeecBcNttt7luk7Z9HjduHGeeeWbRbYrFfOqppwKl\ni2LHud9RXCOb7NRNN9009LbDVOq3oZzPJg3Hqlu8xTL0vO7j1KlTOf744/0Hl7C9994bgBdffLHG\nuqQzi/r37w+UP2HDzjvvDMDAgQOBquuucqT5u5qVjL5yZb1WURjM33W77bZbbpkyi0RERERERERE\nJBXSP51ECV6mVLzzzjuBqrpDhdXg0/S8fxCFWUW1TRZqFUlpdevWrdGXY8eOBbKfWeTnzlZSgt5R\nz+L3zm0mLXNnU9Jn6623TjqEsnn5bpS6i57G71exrKKg9WBKtZFmWYx7vfXc7yVnLaPI1NxzU/g3\nwEknncS9995b1ntkOasInDOKCmdqypr58+cD0KGDcyJFGmeR9MLr98+p5k+aVUpGkdt+9O3bly++\n+MJx3RlnnAHAH//4x8jiCkKZRSIiIiIiIiIikpP5mkXffvstAA0bNizWvuPycvbdzEz1l7/8pYzo\n4pW/P6Y6/rbbbptUOJ4FGU32MqtYlkaki8lCzSITY/fu3QF47LHHArVTKOq+POigg6r9+4UXXgBq\nZiN6Ze423n///Y7r/cz2F7YofgPatWuXquPSq6xnJ7rNIlIs/jZt2gAwd+5cz69JyjHHHAPAAw88\nkFuWxjiDaNSoEQBff/214/o07q+f3/CtttoKgM8//7xk+x07dgTg9ddf9xFdOLbffnsArr322mrL\ne/To4fqaNPaVVzNmzCg6Q2kpH330EVD1uSWlcePGJWuamozfUrWynGS1jw855BAA/v3vf7tuk9S+\nFZ5Ppk6dCpSfvfXmm28CsPvuu1dbXjgTXNaUumYz2bbLli2LI5zASu1PGme/LCaqcRXVLBIRERER\nERERkVSoVZlF5rnlW265xXP7WcgoMvL7csMNNwSqZgZLs6iOwayNrpeSpcyifIMHDwZg0KBBvtsw\nor7TFff5MA13JUeMGAHA+eefH7itNOxPEFnPLDLKmdlMs6Cli9sxeMIJJwAwZcqUOMPxpNh588or\nrwRqZuR4MX78eAB69erlL7AYVOox6Tez6KuvvgKqZipO2t577+1YjyeorPaxyZ5+5plnXLdJet/6\n9OkDwN///vca63bddVegapayTz75BKjKkB0yZAhQvPZL0vvnV6XVdiv1N/mPP/4IQP369WOLKQg/\nYwzlUGaRiIiIiIiIiIikQq3ILPLDjE5feumlobYbpfy+zMrIMsDQoUMBGDBgQKjtZukz8CILmUVG\n2OeV3//+90DVM+hRiet8mOZjs5zP4Oijjwbgn//8Z1ThxGKPPfYAqmZHMXbccUcAFi5cGHtMQRT2\nodOsL4XbTJo0CYBTTjkl4uj8y4+5S5cuACxatAiAjz/+GEj3d6uYUt+7NO9X2OfNNNQoKqVSM4ry\nlduvo0aNol+/fhFF409Yx2YW/yYA6NSpEwCvvPKKp+3TdAyH1XdZr1EEMHLkSM8zXqepD4vJ8m9e\nMWH+Hs6cOTPs2fp8ZRZpsKiASZm+7rrrQmkvTlkdLCpV0LNcWdr3cmRpsMgIen554403APjDH/4Q\nRjgluRUI9ss8RjF9+nQA7rnnnlDalXBVyuNnhbI2xXUpXvYnC/tR6Pvvv2eTTTZxXJeF/Sn3D1I3\nWdjXSj1XOHnttdeAqsE7N/379wdg2LBhkcdUruuvv57LL7/c9+vT/Pinm3KuY+bNmwfALrvsEmlM\nfrjdxPGqkr6TxX77zCN5m266aVzhBFKpg0SF2rdvD8BVV10FwBFHHOH5tRF+BnoMTURERERERERE\ngsl8ZlHLli0B+OCDD3y//4oVK9hss818vz5pTqPvWRyZ3WCDDQD46aefyn7t2WefzZgxY8IOKVWy\nmFmUb9asWQB06FB8UPuvf/1rJgrKS+Uo/G4tWbIEgG233TaJcEKz8847A1V3j51k6bfC6Xrl559/\nBqp+P7KoUh5pKud6cuDAgQDcdNNNUYUTKjMF/Icffui4Pov9VVuUmzHcrl27VF9XhfF3W6Ucr+a6\n0pxPnn322STDCV2lZdMW258sPH5cAZRZJCIiIiIiIiIiwdRNOoCgTOFRM7K6cOFCWrRo4em1WRqN\nLSZ/FNZMNZhFq1atAiqnX6S6uOoOiQSV9YwiY/78+UDlnFMrZT+MNGR2h6nS+iefqTtRyBT5l/RK\nYz2euHz22WcANGvWDID//e9/SYYTutp8XWlqFWXBI4884rrOFCBXRlF6KbNIRERERERERERyMl+z\nSKosXLgwV8NJKk/hd7WS7+KKxMntd7ASptyV9MrqDKZSVQ/MZO+JiISlfv36AKxcudJ1myz9ZhQb\na8jSflQA1SwSEREREREREZFgMl+zSKooq6iyafRdJB7PP/88AAceeGDCkUgl0zk9u5RRJCJJuPzy\ny5MOoWz16tUD4Jdffskt0+9fdgTKLLIsq6FlWQ9YlvW+ZVnzLcvqZFlWI8uynrYsa+G6/2Z3TnoR\nERERERERkVomUM0iy7ImAC/atj3Wsqx6wEbAZcA3tm0PsSxrILCZbduXlGhHNYtERERERERERMLl\nq2aR78Eiy7IaAG8B29t5jViWtQDY37btzy3L2hqYbtt2qxJtabBIRERERERERCRcsRe4bg4sB+6y\nLGuOZVljLcuqD2xp2/bn67ZZBmwZ4D1ERERERERERCRGQQaL6gLtgTts294d+AEYmL/Buowjx6wh\ny7L6WJb1hmVZbwSIQUREREREREREQhRksGgpsNS27Znr/v0AawePvlj3+Bnr/vul04tt2x5j23YH\nP+lQIiIiIiIiIiISDd+DRbZtLwOWWJZl6hEdBMwDHgVOW7fsNOCRQBGKiIiIiIiIiEhs6gZ8/Z+B\nSetmQlsMnMHaAagplmX1Aj4Bjg/4HiIiIiIiIiIiEhPfs6GFGoRmQxMRERERERERCVvss6GJiIiI\niIiIiEiFCfoYmmREgwYNAHjllVcAaN26NQCWZSUWk4iIiIiIiIikjzKLREREREREREQkR5lFFc5k\nFK1YsSLhSERERERERERqmjFjBgD77rtvblnXrl0BeOqppxKJqbZTZpGIiIiIiIiIiOQos6hCrb/+\n+oB7RpFqFaXLueeeC8CoUaNyy/r27QvA7bffnkhMSVq4cCEALVq0cFyv41dEnDjdlTR03pA4PfbY\nYwBsueWWAHTo4H0SmmeeeQaAgw8+OPzARDLu22+/BaBhw4aO63Wuzy6n3+7f/va3CUQihjKLRERE\nREREREQkx7JtO+kYsCwr1iDWW2/tGNmaNWtKbnvllVcCcN1110UaU9jc+lWj7clo2bIlAK+//jrg\nfjekHDvssAMAixcvDtxWmixcuNA1o8iolOP46quvBuCqq67KLWvWrBkAS5cuTSIkkUx57bXXAOjY\nsWPR7SrlnJF1J510EgD33ntvbtn48eOBqjvKpc7/kM7+POqoowB4+OGHQ2szjftZKf785z8DcPnl\nlwNVGWB+qJ+iM2HCBAB69uxZ1usGDx4MwKBBg0KPScLTuHFjAL788suS22b5e/bpp5/SpEkToOop\nkn79+hV9zWuvvZa7ttlwww0BWLVqVZAwZtu27T3FdR1lFomIiIiIiIiISE6tyCyqX78+ACtXrgzc\nVtpHNd36s0uXLgA8++yzcYYTm/wR21Li7sOWLVvywQcfFN1m9OjRADz66KOu2zz++OOOy++88076\n9OnjP8CU8XJOSvv30Kti+7r11lsDsGzZsrjCKemggw4CquppFJPG+AvtvffeALz44ouO67fbbjsA\nPvnkk7hCCqRRo0YAfP31155f07t3bwDGjh1bbfkZZ5wBwN133x1OcCEzdfl++eWXotuZu3ImqzON\n8n8jFixYAMBOO+2UZEih22CDDQD46aefQm03Tb8Ffq6nTfzDhg0D4KKLLiq6XZKynrFeTgaDH1n5\nHLIm6N+pWbgWqY2CnC+z6IorruDaa68FvO9H/mdUr149AH799dcgYfjKLKrowSKvF5PPP/987oLZ\nMI+CTJ8+vdryNB+ohx12WI0BBTMIcc455yQRUmSCHLdx96FTrEFicLrgGTJkCACXXnqp73aTVk6f\npvF7GOagNMCcOXMAaN++fSjtlSv/0ckgj00uWrSoWntp4PT4XzFpPN7yPffccwAccMABobedxn33\nO6Ds9rrWrVsDMH/+/GCBlalNmzYAzJ07t8a6NH7uQXjpM1Ma4KuvvgKqJnl48MEHXdtI0+f0n//8\nB6i6fjTMY0433HBDyTbcPqeBAwcCcNNNNwUJMRAvfXjCCScAMGXKlKjD8WTo0KEMGDAglvdK07FY\nScL6OzWLpRvC2vf+/fsDVYPSSfKzT0uWLAFg2223DTucyL300ksAdO7cObes1LnC6Ro1ycEiPYYm\nIiIiIiIiIiI5FZ1Z9N577wFVdw0NMwXfN998U7KNws8nzXcO0n7XLQiTPlpOAUJzl7JOnTqO6+P6\nbL7//ns22WSTyN4zv9+z3N9O+5HmtPcffvgBgI022ijS94lrX48//ngA7r///kjaT0OfgfMxZSYy\nuOuuu4Cqu1j50hK/kyh/x7fbbrvUPYZXbH+L9ZPb637++Weg6lGpuFTyb7YR5mPF+W39+OOPQFVG\nZxoUTp5STl+aR8n//ve/O67/17/+BUC3bt2ChOiLWx+azC9TrNVJ0sdznH/jdO3aFYCnnnoq0veJ\ncp8aNmzId999F1n7fvjNOk/z9WMpxfZ55syZntooNekDJPNZZOnJkDD4+Z0fM2YMAGeddVZumTKL\nREREREREREQkFeomHUAUzNSshRlFhpeMIlMoNAvMdK352rVrl0Ak4Zk1axYAHTp4GwAdPHiw6/SY\nSWfPbbrpphx22GGhtztu3LjQ20y7NNxVSPp4ClunTp0AfxlFW2yxBQDLly+vttzpMzLLkupDp5gu\nvPBCAEaMGFFt+fDhwwH3YrO1yRdffJF0CDlBvnumhmEaXHHFFTWWFZvcwKs0ZULvvPPORdebabDv\nueeekm2tXr26xjKTrZsm//vf/wB/n7tbRpGRREaRm8K+u/3224G1EyAUTn5gjsmNN94YqMrITbOu\nXbuWzA6qtOsAY8WKFa7r0pjNZxTWCQP3DPWkr0WK8XJc7bnnnp7a6tSpExMnTgSgRYsWgeIKk/nc\nTV0zU7PvuuuuA5w/A5O1lyXm6aZ8JhuzlPyMIiNgRlEgyiwSEREREREREZGciqxZ5LZP8+bNA2CX\nXXbx3UYaR6KdaqekMU6v8mv8uCln9q8s9WU5KrneRRr7zM+50mRllFNry9R/MHcgot5nL/u11157\nAfDqq6/6bsvMqJZUPYT82Pbff38AZsyY4fk1af5umRnzTCam+a0zs3A4KdXvJjv17bffDiPEUDjF\nPH78eAB69epV9muNuGsWhX3udpvdb8MNNwRg1apVvtv2a+jQoQA1ZqIqJybTp2PHjq2xLs3fx3KU\n+h4mWavooIMOAqiRLVTOZ1+4f+Vch4fB6fM1tT/M7JHl1BjyWy8tTGn4uw28XxcEFfTc7VaPMY2z\nRYd9fF1//fVA1YyMYbQZtWKZRVHXAwtTkN/5CP++U80iEREREREREREJpiIziww/2Qn77bcfANOn\nTy/7tUnJ389FixYB0LJly6TC8a3YsTh16lSg6g6BF+aueufOnV23SWN/luL0Od16660AXHDBBXGH\nE7rHHnuMP/7xj47r0lTvxk2xGL3MoBbXPpo71dOmTSs7Dj+/G0n1ndN5wM/dnSyeK5yYTKHddtvN\ncb3JbOvXr19sMZUS5G6rqZ1jMhqcxJVZZGp9rFy5stryRx99lCOPPLLs9ho0aAC41xlJwzHbtGlT\nAJYuXVr2a936vU6dOrn6QFlX6lyaxmzacmJavHgxAM2bN/fdRhqYOmPXXnut6zZJ7ZNbFl/QeP7z\nn/8AzvWA8u21116RZhddfPHFANx88825ZcceeywADz74YMnXl5rpNU3HYtiZRWk+v7jJembRP/7x\nDwB69OhRY12pzzvI9apHyiwSEREREREREZFgKjqzyO1ZTVMLxMzgANkcfXXKgurduzeQrZmyvByD\n5Xz+Zna4hx9+uOh2zz//PAceeKDndpNS6q7IggUL2GmnneIMKVJpqsV0+umnA3DXXXe5bnPCCScA\nVTM7FJOmzKLJkycDcOKJJ0b6PgsWLABI7Bh1Op5MzSLDLZM0Xxp/A8p1yCGH8O9//7voNmnaz2XL\nlgHONb/KjXP77bfnww8/dFyXdGYR1J67xsWYzNhbbrml6HZZ2y83V1xxRdFMFUhXZlE5M9i5tWFm\n/fOTSZcEr38jVcoxWczvfvc7AD7++OMa69K8/7U1s2i//fYreW2Tpn03sp5ZFCQjM4a/f5RZJCIi\nIiIiIiIiwdRNOoAoDRo0CKiq5/Lll18CcN555wFVmUVpmN3AD6cR4yxlFBWbZcfP5/7EE08AcOih\nhxbdrnv37sDa2jhpZvryzDPPdFx/9tlnAzBmzJjYYvJq9erVwNraEm4K7xQUu5uQlGIZRSeffDLg\nLaPIKzNLSxxOOukkILrMoo8++ghILqOoGC+ZRIXM8Znm3wQ35o5wsayiLO5XOUztlCSZzEInplbB\n3nvvXbSNUjWnoOq3IQvq16/vmGlVjG3bmT5ew86mjks5GUVu9SUrLaPommuuiTiS9HDKKIKqOnfi\n36mnnlp0faknJZw4Xeek8bxSKczskX40atQoxEjCV9GPoRUKq0htWjjtz+9//3sAZs+eXfS1W2yx\nBQDLly8PPzCPivXH4YcfXu3fpnC3KeT597//HYCjjz7a8/vdd999QNUfyWlkBrDcCjxDOo/NTp06\nAfDKK6+E0t7LL78MlP7DKWpOx2iQ6WJLnYPq1avHr7/+Wna7YfDyyJ1XaXw0sthnb6ZfN9wGaCGd\n3z83CxcuBKBFixZlv3bHHXcE1v6BkNQxGVexz7h/G6K+7krjMfrnP/8ZgNtuuy2U9tK4j4WyNBFA\nofbt29e4jvQSW5s2bQCYO3duteVZuP4yvPRbw4YNAfjuu++iDidxWX/cNQvxm8GiiRMnOq4vJ0bz\nd93mm28eqJ2kZPUxtLAfPwN44YUXciVnQqLH0EREREREREREJBhlFhXIwqirmYp2yZIlgdsaMmQI\nl156aeB2/Ijr2DvjjDMAuPvuu2N5Pz+yemx6iftf//oXUDxbKt/LL7+cyowiw08/lMoY++9//wvA\npptuWnbbcfD6XTXnpG233TbKcGKz/fbbA1QrjJym76GXR5LCdOWVVwJw3XXXxfJ+YUzdHWV7fpl0\n9WeeeSaS9tNwjAb5fS/MfE7TpAdG2NcvQ4YMAUjseqzQVlttxeeff15tWanPPKslHYysXodFpdTn\nMXr0aADOOeecOMIp2/rrrw/AL7/84rg+TX0ZRvbTZZddBsDgwYMDtZO0SsssMpPgOE1sY/bLrRxE\nBP2lzCIREREREREREQmmVmQWbbXVVgA17pIUxBBlCKEqNVoOVXVVzJTARpMmTQCYNm1ablnS+x7V\nMWiK7vXr1y+S9sNQbN/TPMWs21SkxUb/szAF7dVXXw3AVVdd5bpNFFNct27dGoD58+eX3XbUHnvs\nsZJZYQsWLADSWcw6iLRnFiX1+x31Z+C2X0Eym9KYnWLMmDGDfffd19O2++yzD7C2IHZaMqWMo446\nquxCrP3792fYsGFFt0lT34X9nWvXrh1QfMKPpIS5r2n5rjnxsp9Tp04F3At3V5L3338fgFatWjmu\nN7VR33zzzdhi8uO1114DoGPHjo7r03RMljoG99lnn9wkCH7aSNO+luK0HxtuuCEAq1atijscz8I8\nX0ZYVzj+zCLLsi60LOs9y7LetSxrsmVZG1iW1dyyrJmWZS2yLOt+y7Lim95HREREREREREQCqev3\nhZZlbQOcB7S2bfsny7KmACcChwO32LZ9n2VZo4FewB2hRFsmk4FTLKMIYIcddogjnNAUm5mm1Ohx\nGu8EFMZ8yCGH5P7fZKpcdNFFnttL8wh0qdmJzJTHn332WW6mNzNa3b9/f6Aqi8PUwYlT4cj5xhtv\nDBSfEvrbb7+NNKYwxZ1RZKQ1owiK15r66KOPgMrLKDIKZ/RJmxdeeAHAc1aKX2vWrAGgbl3flwyh\n8JNRdO6550YQSbjCmu3EZF4lxUtW0dZbbw3AsmXLfL/PnDlzfL/WrzRk4cetXr2193qLZbHnO++8\n82rMdpfmqeW99Kk5pv1kFD3xxBMAHHrooUA2sjuKfSb7778/kM6/IwpNnjzZNaOoNmnWrFnSIYTC\n/M307rvvJhyJO/P9Xr16NQB16tTx3VaSM5U7CVqzqC6woWVZdYGNgM+BA4EH1q2fABwV8D1ERERE\nRERERCQmgWoWWZZ1PjAY+Al4CjgfeM227Rbr1jcDnrBtu02JdiK5ZVPOvmVhxL+Qn+dT0/Tsfzm8\n9GX37t2BZDJuSon6ruSuu+4KRDvq3rZtW956661qy4LMjuJm1KhRidWZCvuZ7zBmuEhKsdi/+uor\nABo3bhxXOInIyvkyC/XAyhFmLZ6s9KEfaatZBFUxhZFBlIb9i/q3O22zoAVV+Hml8bvmpU9PPvlk\nYG2GSr7DDjus2r8ff/zxst8/TZ+JU10+Iy0ZpX4U6+Mwzk1RcYvbb82iNB1rXnm5Dp8wYQIAzz//\nPJDu2a6duO1jDPXA4q1ZZFnWZsCRQHOgCVAfOLSM1/exLOsNy7Le8BuDiIiIiIiIiIiEK8hwcRfg\nI9u2lwNYlvUQ0BloaFlWXdu2VwNNgU+dXmzb9hhgzLrXhnrrxm3Ezjw7PXHiRMB5JD3rCp8Pb9u2\nLUCNjBAj7aPOI0eO9LRd9+7dU5lRBPHUOiisrRJmv3bq1AmAV155hcMPPxyoehbf8LKPhbPTdevW\nDag+Mx9A3759Uz2DXSlZrW3RsmVLAD744IOS21Z6RlGlMTMrZp2pZ+c022Ihp+/hnXfeGXpMSalf\nv37SIThK+zWFV08//X9UnDgAABScSURBVHQs7zNw4EAAFi1aBMC4ceNied8opLFOZD6v15MA9957\nb7X/Vprrr78egMsvv7zGuiVLlgCw7bbbxhpTGLxcfxXWsU3zjIRevPfee47Lk6jt5sWYMWMAOOus\ns8p+bWH/9uzZE4C77roLyP7vT1rrgQWpWfQfYE/Lsjay1vbOQcA84Hng2HXbnAY8EixEERERERER\nERGJS9CaRdcAJwCrgTlAb2Ab4D6g0bplp9i2/XOJdkK7Fd+tW7camQp57wPAiBEjADj//PNrrMsi\nP32Y9v116iMnad6PTz9dm1TXpEkTz6/xuz9ux4CZ6W/x4sW+2i1s28x+NmDAAKD47GGw9q7VDTfc\n4Ll9I6l+Lee7ZLL4jjvuOABat27t+bV77bUXAK+++moZ0UWjQYMGAKxYsaLktmn+vkUhTcemk0qr\nVVSo2P79/PPay4rf/OY3ntvL6ufgpH79+qxcudJxXdb386GHHgLIzQhqPPnkk0DNujFRMJlFXbp0\nCdyW6Q+TAVDstyLLfZf/ff3tb38LwDfffJNUODW8//77ALRq1SrROJLs4z322AOAmTNnOq5fsmRJ\nJjKK2rdvD8Ds2bMjab9v377V/n377bdH8j5uzDnO1MRyOmbuuGPtRON/+tOfHNtI67kkyuz7tO6z\nccwxxwDwwAMPVFv+2WefAbDNNttEHYKvmkWBBovCEvVjaIUHjymMZdLXnLbJomJ9+eOPPwLpTV0v\nlOXCwIaX79bZZ58NVKVlBmWmd73//vurLQ/yeZVzjujatSvg7TGRYu0n1b9XX301UHoQzC+3oplJ\nKGeQyMjC9y5MaTo2C33//fdssskmRbdJS6x+hfHH3TPPPAPAwQcfHEpMaeE0WJR0f9944425x6qM\ncmIaNmwYABdddJHj+jj3z8/58dZbbwXgggsucFxfmwaL9tlnH4CSRXnj1KhRIwC+/vrrSN9n3rx5\njsv79+8P1HyMPw777bcfANOnT3dcf9555wHwt7/9La6QAknqb9cdd9wRgIULFyby/vkeeWTtgztH\nHHFEteVZOYdE0Ydp3/cUTNoQb4FrERERERERERGpPNmbDzEAU8A1P6OokqR9RNWrsKcvT6Oo9mPK\nlCkAbLTRRkBV0TfzmYb1vpMmTQKqMqN++OEH322ZmEyMu+66a8Do/IsqsyiNx205d8wlPbzcjUvj\n8ebHTjvtBJRXfN2olM8gzbwci8uXLweqssPMY0lmOcDmm29etI0kfhO+++47oObvk7FgwQKg6hgN\n6uGHHw6lnbRIU0aREeYjceYayEweYK690soto8hkO2UloyiIevXqAXDLLbcANR8188L8Bun3JTjz\nGZoMKXMsFsvacvvNMRPoSDSUWSQiIiIiIiIiIjm1IrOo1N2vNWvWxBSJ+DV69OikQwhku+22A+CT\nTz6J5f3uvvvuav81BcODGDx4MIMGDQrcjps03akpjMXPs9VpLhTZrVu3srZPU9/UVqbeRm1k7jS6\nZXkAPP/88wAceOCB8QUmJZmsIT91Ykxtn/nz54cakx9RnwP/7//+L9L2pUpt+j1zmw4+jTWlyvHR\nRx8B0Lx582rLS9UNA+jXr1+1/xq1KWs3TY488kjP22b58zc1+fJlZX+UWSQiIiIiIiIiIjkVORua\nUenTClea2lCrSLKrU6dOQNVUyoUZiePHjwdg2bJl8QbmQ7du3Zg2bVrRbUw23znnnBNHSKmUptnQ\nvMzko/Nk7ZGm2dDCmpJcx2+2pel8Kc79EXatLUnOuHHjADjzzDOrLdd3Lp1Scn7UbGgiIiIiIiIi\nIhJMRWcWSTblH5MaIReJhtu5X9+5Kvmf0fDhwwG4+OKLkwoHcO63/fffH4AZM2bEHI0kqfBY0HdX\nkvTcc89xwAEHVFumYzJ+p59+OlA1Gy7AvHnzANhll12SCEkiZH4HhgwZAsCll16aZDiSbsosEhER\nERERERGRYJRZJCIiIiIiklFmltP8eoRz5swBoH379onEJNFThqmUQZlFIiIiIiIiIiISjDKLRERE\nREREMsrMxLrlllsCMHPmTPbcc88kQxKRdPGVWVQ3ikhEREREREQkeltttVXSIYhIBdJjaCIiIiIi\nIiIikqPBIhERERERERERydFgkYiIiIiIiIiI5GiwSEREREREREREcjRYJCIiIiIiIiIiORosEhER\nERERERGRHA0WiYiIiIiIiIhIjgaLREREREREREQkR4NFIiIiIiIiIiKSo8EiERERERERERHJ0WCR\niIiIiIiIiIjkaLBIRERERERERERy6iYdQBJGjhwJQN++fV23GTVqFAD9+vWLJSYvXnrpJQA6d+4M\nQMeOHQF4/fXXE4tJRERERERERCqLMotERERERERERCTHsm076RiwLCuSIMLYt0mTJgFwyimnANCp\nU6fculdffTVw+168//77ALRq1cpxvWVZvtseMWIE559/fmjtiT9ejlX1i4iIiIiIiJRptm3bHcp9\nUcnMIsuyxluW9aVlWe/mLWtkWdbTlmUtXPffzdYttyzLus2yrEWWZb1jWVb7cgMSEREREREREZHk\nlMwssixrX2AlMNG27Tbrlg0FvrFte4hlWQOBzWzbvsSyrMOBPwOHAx2BW23b7lgyiBRnFhUTV6aH\n2340a9YMgKVLl5bd5o033gjAwIEDc8tMDafbb7+97PbS5KGHHgLg6KOPLrpdkpk6QY5NZRiJiIiI\niIiIR9FkFtm2/QLwTcHiI4EJ6/5/AnBU3vKJ9lqvAQ0ty9q63KBERERERERERCQZfmdD29K27c/X\n/f8yYMt1/78NsCRvu6Xrln1OjBYuXFh0/ZNPPgnAggULALjgggsYMWIEQI36PWnmJ6OomKQziuKu\nn3XZZZdxww03xPJeY8aMAeCss84qul3//v1rLLv55pur/Xvy5MkAnHTSSSFFJyIiIiIiIlLF72BR\njm3btp/HyCzL6gP0Cfr+IiIiIiIiIiISHk+zoVmWtR3wWF7NogXA/rZtf77uMbPptm23sizr7+v+\nf3LhdiXaDzWlxEMdJt+vLfX6MAXZDzdDhw4FYMCAAaG0F4ZPP/0UgCZNmsTyfmeccQYAd999dyzv\nV6wfd9xxR6B0NpxTO6pdJCIiIiIiIiVEU7PIxaPAaev+/zTgkbzlPdfNirYn8F2pgSIRERERERER\nEUmPko+hWZY1Gdgf2NyyrKXAVcAQYIplWb2AT4Dj123+OGtnQlsE/AicEUHMkVi1apWn7bKazRF3\nPSAv7rjjDqB4RtHw4cMBuOiii4q29fLLL9O5c+ei28Tdd6a2UL6vvvoKgMaNGwduv2nTpkD4tatE\nRERERESkdis5WGTbtlsV3YMctrWBvkGDEhERERERERGRZHiqWRR5EDHXLDr22GMBePDBBz1tD7Dx\nxhsD8MMPPwSMzju3uAYPHgzAoEGDfLfhJO7Mm2KxlRvLeuutx5o1a0JpKyxOtZiCxFL4eT3//PMA\nHHjggb7bFBERERERkYoWa80iERERERERERGpQCUfQ8siM9vVXXfd5bj+gQce8NzWG2+8AcSbURTE\nE088AcChhx6acCSlXXnllQBce+211Zb7yb5xyioyM40lJT+jqFmzZglGIiIiIiIiIuJdRT6GZgTZ\ntzQUst5jjz0AmDlzpuN6MxhSt+7aMb958+YFer807HO5nPrYDPD94Q9/iDucavJjC/LZduvWDYBp\n06ZVW57F/hIREREREZFY6TE0EREREREREREJpqIzi4w0F3n2Iow+MgW6V65c6bpNGvfdzeLFiwFo\n3rx5jXVp2Y+gmUUzZswAYN9993Vcn5b9FBERERERkdRSZpGIiIiIiIiIiARTkQWuDT8ZOeY1lZK1\nUSn7YVxyySWAc0bRCy+8EHc4RS1ZsgRYW9zaHFem0PXSpUsdX5OGTD8RERERERGp3ZRZJCIiIiIi\nIiIiORVZs6hTp04AvPLKK4HbSkNmzvbbbw/Ahx9+WHS7UaNGMXz4cKCqpk+hYv2dhn0txS3+d955\nh7Zt28YcjTf77bcf06dPL/t1hf1RuO9Z6C8RERERERFJlGoWiYiIiIiIiIhIMBWXWdSgQQNWrFgR\nVnPsuuuuALz77ruhtZmkSs0sSnvse+yxBwBDhgyptvy2224D4J///Kfra7O6zyIiIiIiIpI4ZRaJ\niIiIiIiIiEgwFTcb2j777FNym1K1YPINHDgQgFNOOSVYYBKIWx916dIl5kj8ef311wE48MADE45E\nREREREREpDhlFomIiIiIiIiISE7FZRb16NHDdd3o0aMdl8+cOZOOHTtGFZIEUKqm1rPPPhtTJMl5\n4YUXANh3330TjkRERERERERqg4obLCrmnHPOcVxebKCosCBxJapXr17SIZRNxZ1FREREREREoqHH\n0EREREREREREJKdWZRbtt99+AEyfPt3za959992IokmPX3/9NekQanB7/GzRokUxRyIiIiIiIiJS\nu/x/e/cfYmlVx3H8/WH9FRaampKupdGSbKKriCiJmlJsZdofEhtFZgtrKGRQiNYfUiAYQVa0CYua\nFqbJliVB5aKC/bOapuWvfqxmuIu6U6tWCitr3/54zt69bOnuzNyde+eZ9wsu9znnuXPnuzMfZp49\nc55znFkkSZIkSZKkgexqAeE5KSIZWREXX3wxq1evnvX79HVNnDf6fk/Kv3nTpk0cfvjh//fcpNQ4\nl17ve7YQvxaSJEmSpGl5sKpOmu4HObNIkiRJkiRJA72bWQS73m79jSyU2RqTPMNokmsbB2cWSZIk\nSZJmyJlFkiRJkiRJmp1e7oa2fcbF7s4wuuiii1izZs2eLGleuPrqq8f6+Z1RJEmSJEnS+DmzSJIk\nSZIkSQO9nFm0nbNRXt8kfW3OOOOM1z03SXVKkiRJkrQQOLNIkiRJkiRJA7vcDS3JDcA5wOaqOrb1\nfQP4KPAq8CRwYVW92M5dAawEXgM+X1W/3mURI94NTfPTcBYPPfRQAKampsZVjiRJkiRJ890e2w3t\nRmD5Tn3rgGOr6jjgz8AVAEmWAiuA97aP+V6SRdMtSgtTksFjamrKgSJJkiRJksZgl4NFVXUvsGWn\nvjuraltrrgcWt+PzgFuramtV/RXYAJw8wnolSZIkSZK0B41izaLPAr9sx0cAzwyd29j6JEmSJEmS\nNA/Maje0JF8BtgE3z+BjVwGrZvP5JUmSJEmSNFozHixK8hm6ha/Prh0rE28Cjhx62eLW9z+qag2w\npr2XC1xLkiRJkiRNgBndhpZkOXAZcG5VvTJ06g5gRZJ9kxwNLAHun32ZkiRJkiRJmgu7nFmU5Bbg\nTOCQJBuBK+l2P9sXWJcEYH1Vfa6qHktyG/A43e1pl1TVa7tRx9+Bl9uz1GeHYM7Vb2ZcfWfGtRCY\nc/WdGVffDWf8nTN5g+y4g2y8kjxQVSeNuw5pTzLn6jszrr4z41oIzLn6zoyr70aR8VHshiZJkiRJ\nkqSecLBIkiRJkiRJA5M0WLRm3AVIc8Ccq+/MuPrOjGshMOfqOzOuvpt1xidmzSJJkiRJkiSN3yTN\nLJIkSZIkSdKYTcRgUZLlSf6UZEOSy8ddjzQTSW5IsjnJo0N9ByVZl+Qv7fmtrT9JvtMy/4ckJ46v\ncmn3JDkyyT1JHk/yWJJLW785V28k2S/J/Ul+33L+1dZ/dJL7Wp5/nGSf1r9va29o548aZ/3S7kqy\nKMlDSX7R2mZcvZHk6SSPJHk4yQOtz+sV9UqSA5OsTfLHJE8kOXWUOR/7YFGSRcBq4EPAUuATSZaO\ntyppRm4Elu/UdzlwV1UtAe5qbejyvqQ9VgHXzlGN0mxsA75YVUuBU4BL2s9rc64+2QqcVVXHA8uA\n5UlOAb4OXFNV7wZeAFa2168EXmj917TXSfPBpcATQ20zrr55f1UtG9o+3OsV9c23gV9V1THA8XQ/\n00eW87EPFgEnAxuq6qmqehW4FThvzDVJ01ZV9wJbduo+D7ipHd8EfGyo/wfVWQ8cmOTtc1OpNDNV\n9WxV/a4d/4vuF9IRmHP1SMvrv1tz7/Yo4CxgbevfOefb878WODtJ5qhcaUaSLAY+AlzX2sGMq/+8\nXlFvJDkAOB24HqCqXq2qFxlhzidhsOgI4Jmh9sbWJ/XBYVX1bDt+DjisHZt7zWvtNoQTgPsw5+qZ\ndnvOw8BmYB3wJPBiVW1rLxnO8iDn7fxLwMFzW7E0bd8CLgP+09oHY8bVLwXcmeTBJKtan9cr6pOj\ngSng++2W4uuS7M8Icz4Jg0XSglDd1oNuP6h5L8mbgZ8AX6iqfw6fM+fqg6p6raqWAYvpZkAfM+aS\npJFJcg6wuaoeHHct0h50WlWdSHfrzSVJTh8+6fWKemAv4ETg2qo6AXiZHbecAbPP+SQMFm0Cjhxq\nL259Uh88v316X3ve3PrNvealJHvTDRTdXFU/bd3mXL3UpnPfA5xKN117r3ZqOMuDnLfzBwD/mONS\npel4H3Bukqfpln84i27dCzOu3qiqTe15M3A73cC/1yvqk43Axqq6r7XX0g0ejSznkzBY9FtgSduB\nYR9gBXDHmGuSRuUO4IJ2fAHw86H+T7dV6U8BXhqaLihNpLZGxfXAE1X1zaFT5ly9keRtSQ5sx28C\nPkC3Ptc9wPntZTvnfHv+zwfubn/JkyZSVV1RVYur6ii66+67q+qTmHH1RJL9k7xl+zHwQeBRvF5R\nj1TVc8AzSd7Tus4GHmeEOc8k/KxP8mG6e6cXATdU1VVjLkmatiS3AGcChwDPA1cCPwNuA94B/A34\neFVtaf/p/i7d7mmvABdW1QPjqFvaXUlOA34DPMKOdS6+TLdukTlXLyQ5jm5ByEV0f1S7raq+luRd\ndLMwDgIeAj5VVVuT7Af8kG4Nry3Aiqp6ajzVS9OT5EzgS1V1jhlXX7Qs396aewE/qqqrkhyM1yvq\nkSTL6DYq2Ad4CriQdu3CCHI+EYNFkiRJkiRJmgyTcBuaJEmSJEmSJoSDRZIkSZIkSRpwsEiSJEmS\nJEkDDhZJkiRJkiRpwMEiSZIkSZIkDThYJEmSJEmSpAEHiyRJkiRJkjTgYJEkSZIkSZIG/gtZZWiB\nArGf9wAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 1440x1440 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rM-yh1hhMR8v",
        "colab_type": "text"
      },
      "source": [
        "## Checking if cuda is available"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CcBdIfHC_RNR",
        "colab_type": "code",
        "outputId": "c22df1a4-ad49-47a8-e295-061c5e40a29d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "train_on_gpu = torch.cuda.is_available()\n",
        "\n",
        "if not train_on_gpu:\n",
        "    print('CUDA is not available.  Training on CPU ...')\n",
        "else:\n",
        "    print('CUDA is available!  Training on GPU ...')\n",
        "    "
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "CUDA is not available.  Training on CPU ...\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DLGiRvRRMWBb",
        "colab_type": "text"
      },
      "source": [
        "## Defining model architechture"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "j_KoZlDT_TOk",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# define the model CNN architechture\n",
        "class Net(nn.Module):\n",
        "    def __init__(self):\n",
        "        super(Net, self).__init__()\n",
        "        self.fc1 = nn.Linear(784, 500)\n",
        "        self.fc2 = nn.Linear(500, 10)\n",
        "\n",
        "    def forward(self, x):\n",
        "        x = x.view(-1, 784)\n",
        "        x = self.fc1(x)\n",
        "        x = F.relu(x)\n",
        "        x = self.fc2(x)\n",
        "        return x\n",
        "      \n",
        "        "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "eKPAwyD0Maq8",
        "colab_type": "text"
      },
      "source": [
        "## Defining optimizer"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "W8feIlfC_WPI",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "model = Net().to(device)\n",
        "optimizer = optim.SGD(model.parameters(), lr=args.lr)\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EBeBasBWMd4l",
        "colab_type": "text"
      },
      "source": [
        "## Defining Training loop and validation loop\n",
        "### federated learning training"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EEPpe0uO_WiV",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def train_and_test(e):\n",
        "    epochs = e\n",
        "    train_losses , test_losses = [] , []\n",
        "    valid_loss_min = np.Inf \n",
        "    model.train()\n",
        "    print(\"training started...\")\n",
        "    for epoch in range(epochs):\n",
        "      running_loss = 0\n",
        "      batch = 0\n",
        "      #scheduler.step()\n",
        "      for images , labels in federated_train_loader:\n",
        "        model.send(images.location)\n",
        "        images, labels = images.to(device), labels.to(device)\n",
        "        optimizer.zero_grad()\n",
        "        outputs = model(images)\n",
        "        outputs = F.log_softmax(outputs, dim=1)\n",
        "        loss = F.nll_loss(outputs, labels)\n",
        "        loss.backward()\n",
        "        optimizer.step()\n",
        "        model.get()\n",
        "        running_loss += loss.get().item()\n",
        "        batch += 1\n",
        "        if batch % 200 == 0:\n",
        "          print(f\" epoch {epoch + 1} batch {batch} completed\")\n",
        "      test_loss = 0\n",
        "      accuracy = 0\n",
        "      with torch.no_grad():\n",
        "        model.eval() \n",
        "        for images , labels in validloader:\n",
        "          images, labels = images.to(device), labels.to(device)\n",
        "          logps = model(images) \n",
        "          logps = F.log_softmax(logps, dim=1)\n",
        "          test_loss +=  F.nll_loss(logps, labels)\n",
        "          ps = torch.exp(logps)\n",
        "          top_p , top_class = ps.topk(1,dim=1)\n",
        "          equals = top_class == labels.view(*top_class.shape)\n",
        "          accuracy += torch.mean(equals.type(torch.FloatTensor))\n",
        "      train_losses.append(running_loss/len(federated_train_loader))\n",
        "      test_losses.append(test_loss/len(validloader))\n",
        "      print(\"Epoch: {}/{}.. \".format(epoch+1, epochs),\"Training Loss: {:.3f}.. \".format(running_loss/len(federated_train_loader)),\"Valid Loss: {:.3f}.. \".format(test_loss/len(validloader)),\n",
        "        \"Valid Accuracy: {:.3f}\".format(accuracy/len(validloader)))\n",
        "      model.train()\n",
        "      if test_loss/len(validloader) <= valid_loss_min:\n",
        "        print('Validation loss decreased ({:.6f} --> {:.6f}).  Saving model ...'.format(valid_loss_min,test_loss/len(validloader))) \n",
        "        valid_loss_min = test_loss/len(validloader)   \n",
        "    print(\"Training completed\")    \n",
        "    return train_losses, test_losses"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lSQSmHvy_aE1",
        "colab_type": "code",
        "outputId": "a368a850-60c7-4307-e141-edce5b2a1144",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 580
        }
      },
      "source": [
        "train_losses, test_losses = train_and_test(5)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "training started...\n",
            " epoch 1 batch 200 completed\n",
            " epoch 1 batch 400 completed\n",
            " epoch 1 batch 600 completed\n",
            " epoch 1 batch 800 completed\n",
            "Epoch: 1/5..  Training Loss: 0.542..  Valid Loss: 0.304..  Valid Accuracy: 0.915\n",
            "Validation loss decreased (inf --> 0.304273).  Saving model ...\n",
            " epoch 2 batch 200 completed\n",
            " epoch 2 batch 400 completed\n",
            " epoch 2 batch 600 completed\n",
            " epoch 2 batch 800 completed\n",
            "Epoch: 2/5..  Training Loss: 0.286..  Valid Loss: 0.246..  Valid Accuracy: 0.930\n",
            "Validation loss decreased (0.304273 --> 0.246335).  Saving model ...\n",
            " epoch 3 batch 200 completed\n",
            " epoch 3 batch 400 completed\n",
            " epoch 3 batch 600 completed\n",
            " epoch 3 batch 800 completed\n",
            "Epoch: 3/5..  Training Loss: 0.237..  Valid Loss: 0.212..  Valid Accuracy: 0.939\n",
            "Validation loss decreased (0.246335 --> 0.212362).  Saving model ...\n",
            " epoch 4 batch 200 completed\n",
            " epoch 4 batch 400 completed\n",
            " epoch 4 batch 600 completed\n",
            " epoch 4 batch 800 completed\n",
            "Epoch: 4/5..  Training Loss: 0.204..  Valid Loss: 0.188..  Valid Accuracy: 0.946\n",
            "Validation loss decreased (0.212362 --> 0.188346).  Saving model ...\n",
            " epoch 5 batch 200 completed\n",
            " epoch 5 batch 400 completed\n",
            " epoch 5 batch 600 completed\n",
            " epoch 5 batch 800 completed\n",
            "Epoch: 5/5..  Training Loss: 0.180..  Valid Loss: 0.169..  Valid Accuracy: 0.951\n",
            "Validation loss decreased (0.188346 --> 0.169152).  Saving model ...\n",
            "Training completed\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xLDOpNGoMqYS",
        "colab_type": "text"
      },
      "source": [
        "## defining private test loader for encrypting dataset"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_TRyudgc_cNa",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "private_test_loader = []\n",
        "for data, target in validloader:\n",
        "    private_test_loader.append((\n",
        "        data.fix_prec().share(alice, bob, crypto_provider=crypto_provider),\n",
        "        target.fix_prec().share(alice, bob, crypto_provider=crypto_provider)\n",
        "    ))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "JbP0nWAiMyB0",
        "colab_type": "text"
      },
      "source": [
        "## Encrypting model "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5VcKaC2Z_dmR",
        "colab_type": "code",
        "outputId": "1ac2c4e8-7a4b-4443-c6ac-a968b67f6241",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 87
        }
      },
      "source": [
        "model.fix_precision().share(alice, bob, crypto_provider=crypto_provider)\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Net(\n",
              "  (fc1): Linear(in_features=784, out_features=500, bias=True)\n",
              "  (fc2): Linear(in_features=500, out_features=10, bias=True)\n",
              ")"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wQHWdGUGM67c",
        "colab_type": "text"
      },
      "source": [
        "## Testing on encrypted dataset and encrypted model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_MxC5aapFq0Q",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def test(args, model, test_loader):\n",
        "    model.eval()\n",
        "    n_correct_priv = 0\n",
        "    n_total = 0\n",
        "    with torch.no_grad():\n",
        "        for data, target in test_loader:\n",
        "            output = model(data)            \n",
        "            pred = output.argmax(dim=1) \n",
        "            n_correct_priv += pred.eq(target.view_as(pred)).sum()\n",
        "            n_total += args.test_batch_size\n",
        "            \n",
        "            n_correct = n_correct_priv.copy().get().float_precision().long().item()\n",
        "    \n",
        "            print('Test set: Accuracy: {}/{} ({:.0f}%)'.format(\n",
        "                n_correct, n_total,\n",
        "                100. * n_correct / n_total))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dueN0xrMxNgl",
        "colab_type": "code",
        "outputId": "2d813f27-b114-4242-b5c5-97a1b3a71a63",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "test(args, model, private_test_loader)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Test set: Accuracy: 61/64 (95%)\n",
            "Test set: Accuracy: 124/128 (97%)\n",
            "Test set: Accuracy: 185/192 (96%)\n",
            "Test set: Accuracy: 246/256 (96%)\n",
            "Test set: Accuracy: 304/320 (95%)\n",
            "Test set: Accuracy: 364/384 (95%)\n",
            "Test set: Accuracy: 424/448 (95%)\n",
            "Test set: Accuracy: 482/512 (94%)\n",
            "Test set: Accuracy: 542/576 (94%)\n",
            "Test set: Accuracy: 603/640 (94%)\n",
            "Test set: Accuracy: 664/704 (94%)\n",
            "Test set: Accuracy: 725/768 (94%)\n",
            "Test set: Accuracy: 786/832 (94%)\n",
            "Test set: Accuracy: 843/896 (94%)\n",
            "Test set: Accuracy: 903/960 (94%)\n",
            "Test set: Accuracy: 964/1024 (94%)\n",
            "Test set: Accuracy: 1025/1088 (94%)\n",
            "Test set: Accuracy: 1083/1152 (94%)\n",
            "Test set: Accuracy: 1147/1216 (94%)\n",
            "Test set: Accuracy: 1210/1280 (95%)\n",
            "Test set: Accuracy: 1270/1344 (94%)\n",
            "Test set: Accuracy: 1330/1408 (94%)\n",
            "Test set: Accuracy: 1389/1472 (94%)\n",
            "Test set: Accuracy: 1452/1536 (95%)\n",
            "Test set: Accuracy: 1514/1600 (95%)\n",
            "Test set: Accuracy: 1578/1664 (95%)\n",
            "Test set: Accuracy: 1635/1728 (95%)\n",
            "Test set: Accuracy: 1698/1792 (95%)\n",
            "Test set: Accuracy: 1757/1856 (95%)\n",
            "Test set: Accuracy: 1821/1920 (95%)\n",
            "Test set: Accuracy: 1882/1984 (95%)\n",
            "Test set: Accuracy: 1942/2048 (95%)\n",
            "Test set: Accuracy: 2003/2112 (95%)\n",
            "Test set: Accuracy: 2066/2176 (95%)\n",
            "Test set: Accuracy: 2125/2240 (95%)\n",
            "Test set: Accuracy: 2185/2304 (95%)\n",
            "Test set: Accuracy: 2244/2368 (95%)\n",
            "Test set: Accuracy: 2305/2432 (95%)\n",
            "Test set: Accuracy: 2365/2496 (95%)\n",
            "Test set: Accuracy: 2425/2560 (95%)\n",
            "Test set: Accuracy: 2486/2624 (95%)\n",
            "Test set: Accuracy: 2545/2688 (95%)\n",
            "Test set: Accuracy: 2609/2752 (95%)\n",
            "Test set: Accuracy: 2670/2816 (95%)\n",
            "Test set: Accuracy: 2730/2880 (95%)\n",
            "Test set: Accuracy: 2787/2944 (95%)\n",
            "Test set: Accuracy: 2848/3008 (95%)\n",
            "Test set: Accuracy: 2910/3072 (95%)\n",
            "Test set: Accuracy: 2972/3136 (95%)\n",
            "Test set: Accuracy: 3033/3200 (95%)\n",
            "Test set: Accuracy: 3095/3264 (95%)\n",
            "Test set: Accuracy: 3158/3328 (95%)\n",
            "Test set: Accuracy: 3219/3392 (95%)\n",
            "Test set: Accuracy: 3282/3456 (95%)\n",
            "Test set: Accuracy: 3340/3520 (95%)\n",
            "Test set: Accuracy: 3402/3584 (95%)\n",
            "Test set: Accuracy: 3461/3648 (95%)\n",
            "Test set: Accuracy: 3523/3712 (95%)\n",
            "Test set: Accuracy: 3583/3776 (95%)\n",
            "Test set: Accuracy: 3644/3840 (95%)\n",
            "Test set: Accuracy: 3706/3904 (95%)\n",
            "Test set: Accuracy: 3766/3968 (95%)\n",
            "Test set: Accuracy: 3826/4032 (95%)\n",
            "Test set: Accuracy: 3888/4096 (95%)\n",
            "Test set: Accuracy: 3949/4160 (95%)\n",
            "Test set: Accuracy: 4008/4224 (95%)\n",
            "Test set: Accuracy: 4065/4288 (95%)\n",
            "Test set: Accuracy: 4128/4352 (95%)\n",
            "Test set: Accuracy: 4190/4416 (95%)\n",
            "Test set: Accuracy: 4252/4480 (95%)\n",
            "Test set: Accuracy: 4313/4544 (95%)\n",
            "Test set: Accuracy: 4372/4608 (95%)\n",
            "Test set: Accuracy: 4431/4672 (95%)\n",
            "Test set: Accuracy: 4492/4736 (95%)\n",
            "Test set: Accuracy: 4553/4800 (95%)\n",
            "Test set: Accuracy: 4613/4864 (95%)\n",
            "Test set: Accuracy: 4672/4928 (95%)\n",
            "Test set: Accuracy: 4733/4992 (95%)\n",
            "Test set: Accuracy: 4794/5056 (95%)\n",
            "Test set: Accuracy: 4855/5120 (95%)\n",
            "Test set: Accuracy: 4916/5184 (95%)\n",
            "Test set: Accuracy: 4979/5248 (95%)\n",
            "Test set: Accuracy: 5035/5312 (95%)\n",
            "Test set: Accuracy: 5095/5376 (95%)\n",
            "Test set: Accuracy: 5156/5440 (95%)\n",
            "Test set: Accuracy: 5213/5504 (95%)\n",
            "Test set: Accuracy: 5271/5568 (95%)\n",
            "Test set: Accuracy: 5331/5632 (95%)\n",
            "Test set: Accuracy: 5394/5696 (95%)\n",
            "Test set: Accuracy: 5456/5760 (95%)\n",
            "Test set: Accuracy: 5516/5824 (95%)\n",
            "Test set: Accuracy: 5575/5888 (95%)\n",
            "Test set: Accuracy: 5637/5952 (95%)\n",
            "Test set: Accuracy: 5700/6016 (95%)\n",
            "Test set: Accuracy: 5764/6080 (95%)\n",
            "Test set: Accuracy: 5827/6144 (95%)\n",
            "Test set: Accuracy: 5889/6208 (95%)\n",
            "Test set: Accuracy: 5951/6272 (95%)\n",
            "Test set: Accuracy: 6010/6336 (95%)\n",
            "Test set: Accuracy: 6068/6400 (95%)\n",
            "Test set: Accuracy: 6130/6464 (95%)\n",
            "Test set: Accuracy: 6191/6528 (95%)\n",
            "Test set: Accuracy: 6254/6592 (95%)\n",
            "Test set: Accuracy: 6313/6656 (95%)\n",
            "Test set: Accuracy: 6372/6720 (95%)\n",
            "Test set: Accuracy: 6434/6784 (95%)\n",
            "Test set: Accuracy: 6493/6848 (95%)\n",
            "Test set: Accuracy: 6552/6912 (95%)\n",
            "Test set: Accuracy: 6612/6976 (95%)\n",
            "Test set: Accuracy: 6673/7040 (95%)\n",
            "Test set: Accuracy: 6733/7104 (95%)\n",
            "Test set: Accuracy: 6792/7168 (95%)\n",
            "Test set: Accuracy: 6854/7232 (95%)\n",
            "Test set: Accuracy: 6913/7296 (95%)\n",
            "Test set: Accuracy: 6974/7360 (95%)\n",
            "Test set: Accuracy: 7035/7424 (95%)\n",
            "Test set: Accuracy: 7097/7488 (95%)\n",
            "Test set: Accuracy: 7160/7552 (95%)\n",
            "Test set: Accuracy: 7224/7616 (95%)\n",
            "Test set: Accuracy: 7285/7680 (95%)\n",
            "Test set: Accuracy: 7347/7744 (95%)\n",
            "Test set: Accuracy: 7407/7808 (95%)\n",
            "Test set: Accuracy: 7468/7872 (95%)\n",
            "Test set: Accuracy: 7530/7936 (95%)\n",
            "Test set: Accuracy: 7591/8000 (95%)\n",
            "Test set: Accuracy: 7652/8064 (95%)\n",
            "Test set: Accuracy: 7716/8128 (95%)\n",
            "Test set: Accuracy: 7777/8192 (95%)\n",
            "Test set: Accuracy: 7839/8256 (95%)\n",
            "Test set: Accuracy: 7900/8320 (95%)\n",
            "Test set: Accuracy: 7960/8384 (95%)\n",
            "Test set: Accuracy: 8023/8448 (95%)\n",
            "Test set: Accuracy: 8084/8512 (95%)\n",
            "Test set: Accuracy: 8148/8576 (95%)\n",
            "Test set: Accuracy: 8208/8640 (95%)\n",
            "Test set: Accuracy: 8268/8704 (95%)\n",
            "Test set: Accuracy: 8330/8768 (95%)\n",
            "Test set: Accuracy: 8392/8832 (95%)\n",
            "Test set: Accuracy: 8454/8896 (95%)\n",
            "Test set: Accuracy: 8514/8960 (95%)\n",
            "Test set: Accuracy: 8572/9024 (95%)\n",
            "Test set: Accuracy: 8633/9088 (95%)\n",
            "Test set: Accuracy: 8696/9152 (95%)\n",
            "Test set: Accuracy: 8758/9216 (95%)\n",
            "Test set: Accuracy: 8818/9280 (95%)\n",
            "Test set: Accuracy: 8875/9344 (95%)\n",
            "Test set: Accuracy: 8937/9408 (95%)\n",
            "Test set: Accuracy: 9001/9472 (95%)\n",
            "Test set: Accuracy: 9061/9536 (95%)\n",
            "Test set: Accuracy: 9125/9600 (95%)\n",
            "Test set: Accuracy: 9187/9664 (95%)\n",
            "Test set: Accuracy: 9248/9728 (95%)\n",
            "Test set: Accuracy: 9311/9792 (95%)\n",
            "Test set: Accuracy: 9374/9856 (95%)\n",
            "Test set: Accuracy: 9434/9920 (95%)\n",
            "Test set: Accuracy: 9495/9984 (95%)\n",
            "Test set: Accuracy: 9510/10048 (95%)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "bOtrmSn8NAvu",
        "colab_type": "text"
      },
      "source": [
        "## Wooh ! We achived great accuracy while keeping data and model encrypted !"
      ]
    }
  ]
}