PySyft_Keystone_Secure_DL_CIFAR10_with_Linear.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.6.6"
    },
    "colab": {
      "name": "PySyft Keystone Secure DL CIFAR10 with Linear.ipynb",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/Berenice2018/DeepLearning/blob/master/PySyft_Keystone_Secure_DL_CIFAR10_with_Linear.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Sf9WwGD-uu5Q",
        "colab_type": "text"
      },
      "source": [
        "(notebook copy from Kaggle)\n",
        "\n",
        "**Issues with PySyft: (state of 2019, August 20th)** \n",
        "\n",
        "1.   Unfortunately, PySyft does not support CUDA. https://github.com/OpenMined/PySyft/issues/1893\n",
        "That is why I decided for a simple and small dataset. \n",
        "2.   https://github.com/OpenMined/PySyft/issues/2425\n",
        "3.https://github.com/OpenMined/PySyft/issues/2426 \n",
        "4. \"AttributeError: 'Linear' object has no attribute 'fix_prec'\"\n",
        "https://github.com/OpenMined/PySyft/pull/2364\n",
        "5. Running PySyft code often ends up with an empty AssertionError. It is related to the installation process, because `pip install` could not install succesfully. The current PySyft version does not fit with the current PyTorch, TorchVision versions. \n",
        "6. PySyft currenlty only supports exactly 2 workers, neither more no less. \n",
        "7. We get error messages, when we train with non-linear models, for example while using Convolutional and BatchNorm layers\n",
        "\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "xA0pO5YODqCl",
        "colab_type": "text"
      },
      "source": [
        "The server has access to some data to first train its model.\n",
        "The training is done locally.\n",
        "\n",
        "\n",
        "The end device has some data and uses the server's model to get predictions an its data.  Its data is shared in an encrypted way across the workers: \n",
        "```(data.fix_precision().share(ada, bob, crypto_provider= crypto_provdr)```\n",
        "\n",
        "The end device receives the current model from the server, i.e. the model is secret shared with the end devices: \n",
        "``` model.fix_precision().share(ada,bob, crypto_provider=crypto_provdr)```\n",
        "\n",
        "Only the final score is decrypted from the server side to verify predictions were on average good.\n",
        "\n",
        "\n",
        "All the training data remains on the end devices. The server model, the inputs, the model outputs etc. will be encrypted as well. \n",
        "\n",
        "The weights of the server's model are not leaked to the client/ end device. The server has no information about the data input nor the classification output."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "G2SWovhrtGBB",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "!pip install syft"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
        "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5",
        "trusted": true,
        "id": "q1DbkeJZxdAq",
        "colab_type": "code",
        "outputId": "2bddc5f4-b803-4129-8862-183eda66f391",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 105
        }
      },
      "source": [
        "\n",
        "import numpy as np # linear algebra\n",
        "#import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n",
        "from glob import glob\n",
        "from PIL import Image\n",
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline\n",
        "\n",
        "import torch\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F\n",
        "import torch.optim as optim\n",
        "import torchvision\n",
        "from torchvision import datasets, transforms\n",
        "from torch.utils.data.sampler import SubsetRandomSampler\n",
        "\n",
        "import syft as sy\n",
        "\n",
        "import os\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING: Logging before flag parsing goes to stderr.\n",
            "W0827 12:56:11.471678 140100619605888 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",
            "W0827 12:56:11.488156 140100619605888 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": "code",
      "metadata": {
        "trusted": true,
        "id": "dbYGqMYoxdAu",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Visualize plot\n",
        "def plot_loss_acc(n_epochs, train_losses, valid_losses, valid_accuracies):\n",
        "    fig, (ax1, ax2) = plt.subplots(figsize=(14,6), ncols=2)\n",
        "    ax1.plot(valid_losses, label='Validation loss')\n",
        "    ax1.plot(train_losses, label='Training loss')\n",
        "    ax1.legend(frameon=False)\n",
        "    ax1.set_xlabel('Epochs')\n",
        "    ax1.set_ylabel('Loss')\n",
        "    #x_ticks = [x for x in range(0,n_epochs,2)]\n",
        "    #plt.xticks(x_ticks)\n",
        "    \n",
        "    ax2.plot(valid_accuracies, label = 'Validation accuracy')\n",
        "    ax2.legend(frameon=False)\n",
        "    ax2.set_xlabel('Epochs')\n",
        "    \n",
        "    plt.tight_layout()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "_cell_guid": "79c7e3d0-c299-4dcb-8224-4455121ee9b0",
        "_uuid": "d629ff2d2480ee46fbb7e2d37f6b5fab8052498a",
        "trusted": true,
        "id": "hxADDCgNxdAw",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# create workers, \n",
        "hook = sy.TorchHook(torch)\n",
        "\n",
        "ada = sy.VirtualWorker(hook, 'ada')\n",
        "bob = sy.VirtualWorker(hook, 'bob')\n",
        "#cyd = sy.VirtualWorker(hook, 'cyd')\n",
        "\n",
        "client = sy.VirtualWorker(hook, 'client')\n",
        "crypto_provdr = sy.VirtualWorker(hook, 'crypto_provdr') #gives crypto primitives we may need"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "trusted": true,
        "id": "WXENc3QoxdAy",
        "colab_type": "code",
        "outputId": "f12b27be-259c-46a4-cc4b-8075f9c242d1",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 105
        }
      },
      "source": [
        "# define the transform\n",
        "transform = transforms.Compose([\n",
        "    transforms.ToTensor(),\n",
        "    transforms.Normalize((0.5,0.5,0.5 ), (0.5,0.5,0.5 ))\n",
        "])\n",
        "\n",
        "# load the datasets\n",
        "fulltrainset = datasets.CIFAR10(root='./CIFAR10_data', train=True, download=True, transform=transform)\n",
        "testset = datasets.CIFAR10('~/.pytorch/CIFAR10_data', download=True, train=False, transform=transform)\n",
        "\n",
        "train_size = int(len(fulltrainset)* 0.8)\n",
        "valid_size = len(fulltrainset) - train_size\n",
        "\n",
        "# split the dataset\n",
        "trainset, validationset = torch.utils.data.random_split(fulltrainset, [train_size, valid_size])\n",
        "trainset = trainset.dataset\n",
        "validationset = validationset.dataset\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\r0it [00:00, ?it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Downloading https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz to ./CIFAR10_data/cifar-10-python.tar.gz\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "170500096it [00:04, 40791182.82it/s]                               \n",
            "0it [00:00, ?it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Downloading https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz to /root/.pytorch/CIFAR10_data/cifar-10-python.tar.gz\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "170500096it [00:04, 41399959.28it/s]                               \n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "trusted": true,
        "id": "azt73D0ixdA0",
        "colab_type": "code",
        "outputId": "7e89212f-b151-4313-a896-76f4bd4ce0df",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "# we assume that the server has access to some data to first train its model\n",
        "train_loader = torch.utils.data.DataLoader(trainset, batch_size=64, shuffle=True)\n",
        "\n",
        "# The client has some data and gets predictions on it using the server's model. \n",
        "# The client encrypts its data by sharing it additively across the workers ada, bob.\n",
        "valid_loader = torch.utils.data.DataLoader(validationset, batch_size=64, shuffle=True)\n",
        "\n",
        "private_valid_loader = []\n",
        "for data, target in valid_loader:\n",
        "    private_valid_loader.append(\n",
        "        (data.fix_precision().share(ada, bob, crypto_provider= crypto_provdr),\n",
        "        target.fix_precision().share(ada, bob, crypto_provider=crypto_provdr)\n",
        "        ))\n",
        "\n",
        "    \n",
        "test_loader = torch.utils.data.DataLoader(\n",
        "    testset, batch_size=64, shuffle=True)\n",
        "\n",
        "\n",
        "len_trainloader = len(train_loader)\n",
        "len_validloader = len(valid_loader)\n",
        "print(len_trainloader)\n",
        "\n",
        "# Let's check that our trainloader returns a pointer to a batch, and that transformations are applied\n",
        "#data, labels = next(iter(train_loader))\n",
        "#data"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "782\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "63epObo5xdA2",
        "colab_type": "text"
      },
      "source": [
        "**Architecture**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "trusted": true,
        "id": "ll1P35xoxdA3",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class Model(nn.Module):\n",
        "    def __init__(self):\n",
        "        super(Model, self).__init__()\n",
        "        self.fc1 = nn.Linear(3*32*32, 1024)\n",
        "        self.fc2 = nn.Linear(1024, 10)\n",
        "\n",
        "    def forward(self, x):\n",
        "        x = x.view(-1, 3*32*32)\n",
        "        x = self.fc1(x)\n",
        "        x = F.relu(x)\n",
        "        x = self.fc2(x)\n",
        "        return x # F.log_softmax(x, dim=1) Do not call log_softmax here, it is not supported yet. \n",
        "    \n",
        "    \n",
        "    # (in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, bias=True, padding_mode='zeros')\n",
        "class MyCifarNet(nn.Module):\n",
        "    def __init__(self):\n",
        "        super(MyCifarNet, self).__init__()\n",
        "        self.conv1 = nn.Conv2d(3, 6, 5)\n",
        "        self.pool = nn.MaxPool2d(2, 2)\n",
        "        self.conv2 = nn.Conv2d(6, 16, 5)\n",
        "        self.fc1 = nn.Linear(16 * 5 * 5, 120)\n",
        "        self.fc2 = nn.Linear(120, 84)\n",
        "        self.fc3 = nn.Linear(84, 10)\n",
        "\n",
        "    def forward(self, x):\n",
        "        #print('x {}, size {}'.format(x.shape, x.size(0))) # x torch.Size([64, 1, 28, 28]), size 0\n",
        "        x = self.pool(F.relu(self.conv1(x)))\n",
        "        x = self.pool(F.relu(self.conv2(x)))\n",
        "        x = x.view(-1, 16 * 5 * 5)\n",
        "        x = F.relu(self.fc1(x))\n",
        "        x = F.relu(self.fc2(x))\n",
        "        x = self.fc3(x)\n",
        "        return F.log_softmax(x, dim=1)\n",
        "    \n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1_P1_Ma_xdA4",
        "colab_type": "text"
      },
      "source": [
        "**Functions train, test**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "trusted": true,
        "id": "Jelf7lf7xdA5",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def weights_init_normal(m):\n",
        "    '''Takes in a module and initializes all linear layers with weight\n",
        "       values taken from a normal distribution.'''\n",
        "    classname = m.__class__.__name__\n",
        "    # for every Linear layer in a model\n",
        "    if classname.find('Linear') != -1:\n",
        "        n = m.in_features\n",
        "        # m.weight.data shoud be taken from a normal distribution\n",
        "        m.weight.data.normal_(0, 1/np.sqrt(n))\n",
        "        # m.bias.data should be 0\n",
        "        m.bias.data.fill_(0)\n",
        "            \n",
        "\n",
        "def train(args, model, device, train_loader, optimizer, epoch):\n",
        "    # initialize variables to monitor training and validation loss\n",
        "    train_loss = 0.0\n",
        "    correct = 0.0\n",
        "    \n",
        "    model.train()\n",
        "    \n",
        "    for batch_idx, (data, target) in enumerate(train_loader): # <-- now it is a distributed dataset\n",
        "        data, target = data.to(device), target.to(device)\n",
        "        optimizer.zero_grad()\n",
        "        output = model(data)\n",
        "        output = F.log_softmax(output)\n",
        "        \n",
        "        loss = F.nll_loss(output, target)\n",
        "        loss.backward()\n",
        "        optimizer.step()\n",
        "        #print(' \\tTrain. Loss: {:.6f}'.format(current_loss))    \n",
        "\n",
        "        # get the loss per batch and accumulate\n",
        "        train_loss += loss.item()\n",
        "        \n",
        "        \n",
        "    # calculate the average loss per epoch\n",
        "    train_loss = train_loss/len(train_loader)\n",
        "    #print('Epoch: {} \\tTrain. Loss: {:.6f}'.format(epoch, train_loss))    \n",
        "    return train_loss\n",
        "         "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "trusted": true,
        "id": "dznlwy9-xdA7",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def validate(args, model, device, test_loader):\n",
        "    model.eval()\n",
        "    test_loss = 0.\n",
        "    correct = 0.\n",
        "    \n",
        "    with torch.no_grad():\n",
        "        for data, target in test_loader:\n",
        "            data, target = data.to(device), target.to(device)\n",
        "            output = model(data)\n",
        "            output = F.log_softmax(output)\n",
        "            \n",
        "            test_loss += F.nll_loss(output, target, reduction='sum').item() # sum up batch loss\n",
        "            pred = output.argmax(1, keepdim=True) # get the index of the max log-probability \n",
        "            correct += pred.eq(target.view_as(pred)).sum().item()\n",
        "\n",
        "    test_loss /= len(test_loader.dataset)\n",
        "    accuracy = 100. * correct/len(test_loader.dataset)\n",
        "    #print('Epoch: {} \\tValid. Loss: {:.6f}'.format(epoch, test_loss))    \n",
        "    return test_loss, accuracy\n",
        "        "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "trusted": true,
        "id": "ULhlb5y8xdA8",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class Arguments():\n",
        "    def __init__(self):\n",
        "        self.batch_size = 64\n",
        "        self.test_batch_size = 64\n",
        "        self.epochs = 5\n",
        "        self.lr = 0.001\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)\n",
        "\n",
        "device = torch.device(\"cuda\" if use_cuda else \"cpu\")\n",
        "\n",
        "kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {}\n",
        "\n",
        "\n",
        "model = Model() # MyCifarNet()\n",
        "model.apply(weights_init_normal)\n",
        "optimizer = optim.SGD(model.parameters(), lr=args.lr) # TODO momentum is not supported at the moment\n",
        "scheduler = optim.lr_scheduler.MultiStepLR(optimizer, [5, 10, 14], 0.1)\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pU4-pNJOxdA-",
        "colab_type": "text"
      },
      "source": [
        "**Start the training**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "trusted": true,
        "id": "bYWWeC3mxdA_",
        "colab_type": "code",
        "outputId": "3082b6fc-8d07-460d-af64-1e699f4d6cbb",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 665
        }
      },
      "source": [
        "\n",
        "valid_losses = []\n",
        "train_losses = []\n",
        "valid_accuracies = []\n",
        "valid_loss_min = np.Inf\n",
        "    \n",
        "for epoch in range(1, args.epochs + 1):\n",
        "    \n",
        "    # initialize variables to monitor training and validation loss\n",
        "    training_loss = 0.0\n",
        "    training_accuracy = 0.0\n",
        "\n",
        "    if scheduler is not None:\n",
        "        scheduler.step()\n",
        "        \n",
        "    training_loss =  train(args, model, device, train_loader, optimizer, epoch)\n",
        "    validation_loss, validation_accuracy = validate(args, model, device, test_loader)\n",
        "\n",
        "    #if scheduler is not None:\n",
        "      #scheduler.step(validation_loss) # in case of ReduceOnPlateau \n",
        "\n",
        "    ###### print training/validation statistics \n",
        "    train_losses.append(training_loss)\n",
        "    valid_losses.append(validation_loss)\n",
        "    valid_accuracies.append(validation_accuracy)\n",
        "\n",
        "    #hour, minute, second = get_time()\n",
        "    print('Epoch: {} \\tTrain. Loss: {:.6f} \\tValid. Loss: {:.6f} \\t Accur.: {:.6f}'.format(\n",
        "              epoch,\n",
        "              training_loss,\n",
        "              validation_loss,\n",
        "              validation_accuracy ))\n",
        "\n",
        "    ###### TODO: save the model if validation loss has decreased\n",
        "    if validation_loss <= valid_loss_min:\n",
        "        print('Validation loss decreased by {:.6f}'.format(validation_loss - valid_loss_min))\n",
        "        #torch.save(model.state_dict(), save_path)\n",
        "        valid_loss_min = validation_loss\n",
        "    \n",
        "##### visualize\n",
        "plot_loss_acc(args.epochs, train_losses, valid_losses, valid_accuracies)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/syft/frameworks/torch/tensors/interpreters/native.py:338: UserWarning: Implicit dimension choice for log_softmax has been deprecated. Change the call to include dim=X as an argument.\n",
            "  response = eval(cmd)(*args, **kwargs)\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Epoch: 1 \tTrain. Loss: 2.068484 \tValid. Loss: 1.940321 \t Accur.: 32.730000\n",
            "Validation loss decreased by -inf\n",
            "Epoch: 2 \tTrain. Loss: 1.886625 \tValid. Loss: 1.842371 \t Accur.: 36.390000\n",
            "Validation loss decreased by -0.097950\n",
            "Epoch: 3 \tTrain. Loss: 1.810776 \tValid. Loss: 1.786286 \t Accur.: 38.410000\n",
            "Validation loss decreased by -0.056084\n",
            "Epoch: 4 \tTrain. Loss: 1.762386 \tValid. Loss: 1.748039 \t Accur.: 39.510000\n",
            "Validation loss decreased by -0.038247\n",
            "Epoch: 5 \tTrain. Loss: 1.739285 \tValid. Loss: 1.744555 \t Accur.: 39.750000\n",
            "Validation loss decreased by -0.003484\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAA+gAAAGoCAYAAADVZM+hAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 1008x432 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "trusted": true,
        "id": "WA-PNGpGxdBC",
        "colab_type": "code",
        "outputId": "e84f6ef6-bc98-4271-8b5c-26512b18ad2d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 85
        }
      },
      "source": [
        "# As the server, send the model to the workers holding the data. The weights are not disclosed\n",
        "#for param in model.parameters():\n",
        " #   param.requires_grad = True\n",
        "    \n",
        "model.fix_precision().share(ada, bob, crypto_provider=crypto_provdr)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "Model(\n",
              "  (fc1): Linear(in_features=3072, out_features=1024, bias=True)\n",
              "  (fc2): Linear(in_features=1024, out_features=10, bias=True)\n",
              ")"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 72
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "trusted": true,
        "id": "iB3_pEUMxdBF",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# perform the encrypted evaluation. The model weights, the data inputs, the prediction and \n",
        "# the target used for scoring are encrypted!\n",
        "\n",
        "def validate_encrypted(args, model, valid_loader):\n",
        "    model.eval()\n",
        "    n_correct_private = 0\n",
        "    n_total = 0\n",
        "    \n",
        "    with torch.no_grad(): \n",
        "        for data, target in valid_loader:\n",
        "            output = model(data)\n",
        "            \n",
        "            pred = output.argmax(1) # get the index of the max log-probability \n",
        "            n_correct_private += pred.eq(target.view_as(pred)).sum()\n",
        "            \n",
        "            n_total += args.test_batch_size\n",
        "            \n",
        "            #decrypt from the server side the final score of the batch items, \n",
        "            # to verify if the predictions were on average good\n",
        "            n_correct = n_correct_private.copy().get().float_precision().long().item()\n",
        "            \n",
        "            print('Valid. set: Accuracy: {}/{} ({:.0f}%)'.format(\n",
        "                n_correct, n_total,\n",
        "                100. * n_correct / n_total))\n",
        "            \n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "trusted": true,
        "id": "pf00j5tPxdBG",
        "colab_type": "code",
        "outputId": "9e6aee56-1068-4cce-8173-6a2de89d8d79",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "validate_encrypted(args, model, private_valid_loader)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Valid. set: Accuracy: 28/64 (44%)\n",
            "Valid. set: Accuracy: 53/128 (41%)\n",
            "Valid. set: Accuracy: 76/192 (40%)\n",
            "Valid. set: Accuracy: 103/256 (40%)\n",
            "Valid. set: Accuracy: 125/320 (39%)\n",
            "Valid. set: Accuracy: 146/384 (38%)\n",
            "Valid. set: Accuracy: 171/448 (38%)\n",
            "Valid. set: Accuracy: 194/512 (38%)\n",
            "Valid. set: Accuracy: 221/576 (38%)\n",
            "Valid. set: Accuracy: 250/640 (39%)\n",
            "Valid. set: Accuracy: 274/704 (39%)\n",
            "Valid. set: Accuracy: 300/768 (39%)\n",
            "Valid. set: Accuracy: 326/832 (39%)\n",
            "Valid. set: Accuracy: 348/896 (39%)\n",
            "Valid. set: Accuracy: 379/960 (39%)\n",
            "Valid. set: Accuracy: 404/1024 (39%)\n",
            "Valid. set: Accuracy: 429/1088 (39%)\n",
            "Valid. set: Accuracy: 454/1152 (39%)\n",
            "Valid. set: Accuracy: 485/1216 (40%)\n",
            "Valid. set: Accuracy: 509/1280 (40%)\n",
            "Valid. set: Accuracy: 536/1344 (40%)\n",
            "Valid. set: Accuracy: 570/1408 (40%)\n",
            "Valid. set: Accuracy: 599/1472 (41%)\n",
            "Valid. set: Accuracy: 619/1536 (40%)\n",
            "Valid. set: Accuracy: 648/1600 (40%)\n",
            "Valid. set: Accuracy: 674/1664 (41%)\n",
            "Valid. set: Accuracy: 698/1728 (40%)\n",
            "Valid. set: Accuracy: 727/1792 (41%)\n",
            "Valid. set: Accuracy: 760/1856 (41%)\n",
            "Valid. set: Accuracy: 777/1920 (40%)\n",
            "Valid. set: Accuracy: 806/1984 (41%)\n",
            "Valid. set: Accuracy: 830/2048 (41%)\n",
            "Valid. set: Accuracy: 855/2112 (40%)\n",
            "Valid. set: Accuracy: 881/2176 (40%)\n",
            "Valid. set: Accuracy: 899/2240 (40%)\n",
            "Valid. set: Accuracy: 929/2304 (40%)\n",
            "Valid. set: Accuracy: 950/2368 (40%)\n",
            "Valid. set: Accuracy: 978/2432 (40%)\n",
            "Valid. set: Accuracy: 1001/2496 (40%)\n",
            "Valid. set: Accuracy: 1024/2560 (40%)\n",
            "Valid. set: Accuracy: 1047/2624 (40%)\n",
            "Valid. set: Accuracy: 1071/2688 (40%)\n",
            "Valid. set: Accuracy: 1101/2752 (40%)\n",
            "Valid. set: Accuracy: 1127/2816 (40%)\n",
            "Valid. set: Accuracy: 1156/2880 (40%)\n",
            "Valid. set: Accuracy: 1188/2944 (40%)\n",
            "Valid. set: Accuracy: 1217/3008 (40%)\n",
            "Valid. set: Accuracy: 1244/3072 (40%)\n",
            "Valid. set: Accuracy: 1268/3136 (40%)\n",
            "Valid. set: Accuracy: 1290/3200 (40%)\n",
            "Valid. set: Accuracy: 1315/3264 (40%)\n",
            "Valid. set: Accuracy: 1342/3328 (40%)\n",
            "Valid. set: Accuracy: 1365/3392 (40%)\n",
            "Valid. set: Accuracy: 1386/3456 (40%)\n",
            "Valid. set: Accuracy: 1414/3520 (40%)\n",
            "Valid. set: Accuracy: 1443/3584 (40%)\n",
            "Valid. set: Accuracy: 1472/3648 (40%)\n",
            "Valid. set: Accuracy: 1496/3712 (40%)\n",
            "Valid. set: Accuracy: 1523/3776 (40%)\n",
            "Valid. set: Accuracy: 1553/3840 (40%)\n",
            "Valid. set: Accuracy: 1576/3904 (40%)\n",
            "Valid. set: Accuracy: 1601/3968 (40%)\n",
            "Valid. set: Accuracy: 1628/4032 (40%)\n",
            "Valid. set: Accuracy: 1651/4096 (40%)\n",
            "Valid. set: Accuracy: 1673/4160 (40%)\n",
            "Valid. set: Accuracy: 1694/4224 (40%)\n",
            "Valid. set: Accuracy: 1719/4288 (40%)\n",
            "Valid. set: Accuracy: 1746/4352 (40%)\n",
            "Valid. set: Accuracy: 1772/4416 (40%)\n",
            "Valid. set: Accuracy: 1802/4480 (40%)\n",
            "Valid. set: Accuracy: 1826/4544 (40%)\n",
            "Valid. set: Accuracy: 1853/4608 (40%)\n",
            "Valid. set: Accuracy: 1879/4672 (40%)\n",
            "Valid. set: Accuracy: 1908/4736 (40%)\n",
            "Valid. set: Accuracy: 1940/4800 (40%)\n",
            "Valid. set: Accuracy: 1962/4864 (40%)\n",
            "Valid. set: Accuracy: 1990/4928 (40%)\n",
            "Valid. set: Accuracy: 2013/4992 (40%)\n",
            "Valid. set: Accuracy: 2036/5056 (40%)\n",
            "Valid. set: Accuracy: 2057/5120 (40%)\n",
            "Valid. set: Accuracy: 2080/5184 (40%)\n",
            "Valid. set: Accuracy: 2102/5248 (40%)\n",
            "Valid. set: Accuracy: 2122/5312 (40%)\n",
            "Valid. set: Accuracy: 2153/5376 (40%)\n",
            "Valid. set: Accuracy: 2177/5440 (40%)\n",
            "Valid. set: Accuracy: 2203/5504 (40%)\n",
            "Valid. set: Accuracy: 2232/5568 (40%)\n",
            "Valid. set: Accuracy: 2259/5632 (40%)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "trusted": true,
        "id": "cfLBVlAoxdBJ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "\n",
        "print(f'objects of ada= {len(ada._objects)}, bob= {len(bob._objects)}, cyd= {len(cyd._objects)}')\n",
        "\n",
        "#ada.clear_objects()\n",
        "#bob.clear_objects()\n",
        "#cyd.clear_objects()"
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}