ProcessingWeightsGPU.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "ProcessingWeightsGPU.ipynb",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "[View in Colaboratory](https://colab.research.google.com/github/kobybibas/FittingRandomDataToResnetCNN/blob/master/ProcessingWeightsGPU.ipynb)"
      ]
    },
    {
      "metadata": {
        "id": "HuEICF0hw_cO",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 323
        },
        "outputId": "cdd5c6ca-f9b9-451e-febe-f7bfc781ec4a"
      },
      "cell_type": "code",
      "source": [
        "!pip3 install http://download.pytorch.org/whl/cu80/torch-0.3.0.post4-cp36-cp36m-linux_x86_64.whl \n",
        "!pip3 install torchvision\n",
        "import matplotlib.pyplot as plot\n",
        "import os\n",
        "import math\n",
        "import time\n",
        "import copy\n",
        "import numpy as np\n",
        "import torch\n",
        "import torch.optim\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F\n",
        "import torch.backends.cudnn as cudnn\n",
        "import torchvision.datasets as datasets\n",
        "import torchvision.transforms as transforms\n",
        "\n",
        "print(torch.__version__)\n",
        "print(torch.cuda.is_available())"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting torch==0.3.0.post4 from http://download.pytorch.org/whl/cu80/torch-0.3.0.post4-cp36-cp36m-linux_x86_64.whl\n",
            "\u001b[?25l  Downloading http://download.pytorch.org/whl/cu80/torch-0.3.0.post4-cp36-cp36m-linux_x86_64.whl (592.3MB)\n",
            "\u001b[K    92% |█████████████████████████████▋  | 547.8MB 48.6MB/s eta 0:00:01"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "\u001b[K    100% |████████████████████████████████| 592.3MB 35.7MB/s \n",
            "\u001b[?25hRequirement already satisfied: pyyaml in /usr/local/lib/python3.6/dist-packages (from torch==0.3.0.post4) (3.12)\n",
            "Requirement already satisfied: numpy in /usr/local/lib/python3.6/dist-packages (from torch==0.3.0.post4) (1.14.3)\n",
            "Installing collected packages: torch\n",
            "  Found existing installation: torch 0.4.0\n",
            "    Uninstalling torch-0.4.0:\n",
            "      Successfully uninstalled torch-0.4.0\n",
            "Successfully installed torch-0.3.0.post4\n",
            "Requirement already satisfied: torchvision in /usr/local/lib/python3.6/dist-packages (0.2.1)\n",
            "Requirement already satisfied: pillow>=4.1.1 in /usr/local/lib/python3.6/dist-packages (from torchvision) (5.1.0)\n",
            "Requirement already satisfied: numpy in /usr/local/lib/python3.6/dist-packages (from torchvision) (1.14.3)\n",
            "Requirement already satisfied: six in /usr/local/lib/python3.6/dist-packages (from torchvision) (1.11.0)\n",
            "Requirement already satisfied: torch in /usr/local/lib/python3.6/dist-packages (from torchvision) (0.3.0.post4)\n",
            "Requirement already satisfied: pyyaml in /usr/local/lib/python3.6/dist-packages (from torch->torchvision) (3.12)\n",
            "0.3.0.post4\n",
            "True\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "82RcK6cfxLXA",
        "colab_type": "code",
        "colab": {
          "resources": {
            "http://localhost:8080/nbextensions/google.colab/files.js": {
              "data": "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",
              "ok": true,
              "headers": [
                [
                  "content-type",
                  "application/javascript"
                ]
              ],
              "status": 200,
              "status_text": ""
            }
          },
          "base_uri": "https://localhost:8080/",
          "height": 154
        },
        "outputId": "fa10f05d-3444-4184-f8c4-eb8bc7b479b3"
      },
      "cell_type": "code",
      "source": [
        "!ls\n",
        "from google.colab import files\n",
        "uploaded = files.upload() # then browse, select the files. It's then uploaded"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "data\r\n",
            "datalab\r\n",
            "Model_corruptiton_0.0_Acc-tr_99.98 (1).pt\r\n",
            "Model_corruptiton_0.0_Acc-tr_99.98.pt\r\n",
            "Model_corruptiton_1.0_Acc-tr_10.24.pt\r\n",
            "Model_corruptiton_1.0_Acc-tr_99.76 (1).pt\r\n",
            "Model_corruptiton_1.0_Acc-tr_99.76.pt\r\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "\n",
              "     <input type=\"file\" id=\"files-50058d2a-5efd-4a1c-9b62-77d056553c4b\" name=\"files[]\" multiple disabled />\n",
              "     <output id=\"result-50058d2a-5efd-4a1c-9b62-77d056553c4b\">\n",
              "      Upload widget is only available when the cell has been executed in the\n",
              "      current browser session. Please rerun this cell to enable.\n",
              "      </output>\n",
              "      <script src=\"/nbextensions/google.colab/files.js\"></script> "
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "ffY8Geg0w_ck",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "#Define the network\n",
        "class BasicBlock(nn.Module):\n",
        "    def __init__(self, in_planes, out_planes, stride, dropRate=0.0):\n",
        "        super(BasicBlock, self).__init__()\n",
        "        self.bn1 = nn.BatchNorm2d(in_planes)\n",
        "        self.relu1 = nn.ReLU(inplace=True)\n",
        "        self.conv1 = nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,\n",
        "                                padding=1, bias=False)\n",
        "        self.bn2 = nn.BatchNorm2d(out_planes)\n",
        "        self.relu2 = nn.ReLU(inplace=True)\n",
        "        self.conv2 = nn.Conv2d(out_planes, out_planes, kernel_size=3, stride=1,\n",
        "                                padding=1, bias=False)\n",
        "        self.droprate = dropRate\n",
        "        self.equalInOut = (in_planes == out_planes)\n",
        "        self.convShortcut = (not self.equalInOut) and nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride,\n",
        "                                                                padding=0, bias=False) or None\n",
        "\n",
        "    def forward(self, x):\n",
        "        if not self.equalInOut:\n",
        "            x = self.relu1(self.bn1(x))\n",
        "            out = self.conv1(x)\n",
        "        else:\n",
        "            out = self.conv1(self.relu1(self.bn1(x)))\n",
        "\n",
        "        if self.droprate > 0:\n",
        "            out = F.dropout(out, p=self.droprate, training=self.training)\n",
        "        out = self.conv2(self.relu2(self.bn2(out)))\n",
        "        if not self.equalInOut:\n",
        "            return torch.add(self.convShortcut(x), out)\n",
        "        else:\n",
        "            return torch.add(x, out)\n",
        "\n",
        "\n",
        "class NetworkBlock(nn.Module):\n",
        "        def __init__(self, nb_layers, in_planes, out_planes, block, stride, dropRate=0.0):\n",
        "            super(NetworkBlock, self).__init__()\n",
        "            self.layer = self._make_layer(block, in_planes, out_planes, nb_layers, stride, dropRate)\n",
        "\n",
        "        def _make_layer(self, block, in_planes, out_planes, nb_layers, stride, dropRate):\n",
        "            layers = []\n",
        "            for i in range(nb_layers):\n",
        "                layers.append(block(i == 0 and in_planes or out_planes, out_planes, i == 0 and stride or 1, dropRate))\n",
        "            return nn.Sequential(*layers)\n",
        "\n",
        "        def forward(self, x):\n",
        "            return self.layer(x)\n",
        "\n",
        "\n",
        "class WideResNet(nn.Module):\n",
        "    def __init__(self, depth, num_classes, widen_factor=1, drop_rate=0.0, init_scale=1.0):\n",
        "        super(WideResNet, self).__init__()\n",
        "\n",
        "        nChannels = [16, 16 * widen_factor, 32 * widen_factor, 64 * widen_factor]\n",
        "        assert((depth - 4) % 6 == 0)\n",
        "        n = (depth - 4) // 6\n",
        "        block = BasicBlock\n",
        "        # 1st conv before any network block\n",
        "        self.conv1 = nn.Conv2d(3, nChannels[0], kernel_size=3, stride=1,\n",
        "                                padding=1, bias=False)\n",
        "        # 1st block\n",
        "        self.block1 = NetworkBlock(\n",
        "            n, nChannels[0], nChannels[1], block, 1, drop_rate)\n",
        "        # 2nd block\n",
        "        self.block2 = NetworkBlock(n, nChannels[1], nChannels[2], block, 2, drop_rate)\n",
        "        # 3rd block\n",
        "        self.block3 = NetworkBlock(n, nChannels[2], nChannels[3], block, 2, drop_rate)\n",
        "        # global average pooling and classifier\n",
        "        self.bn1 = nn.BatchNorm2d(nChannels[3])\n",
        "        self.relu = nn.ReLU(inplace=True)\n",
        "        self.fc = nn.Linear(nChannels[3], num_classes)\n",
        "        self.nChannels = nChannels[3]\n",
        "\n",
        "        for m in self.modules():\n",
        "            if isinstance(m, nn.Conv2d):\n",
        "                n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels\n",
        "                m.weight.data.normal_(0, init_scale * math.sqrt(2. / n))\n",
        "            elif isinstance(m, nn.BatchNorm2d):\n",
        "                m.weight.data.fill_(1)\n",
        "                m.bias.data.zero_()\n",
        "            elif isinstance(m, nn.Linear):\n",
        "                m.bias.data.zero_()\n",
        "\n",
        "                size = m.weight.size()\n",
        "                fan_out = size[0] # number of rows\n",
        "                fan_in = size[1] # number of columns\n",
        "                variance = math.sqrt(2.0/(fan_in + fan_out))\n",
        "                m.weight.data.normal_(0.0, init_scale * variance)\n",
        "\n",
        "\n",
        "    def forward(self, x):\n",
        "        out = self.forward_repr(x)\n",
        "        return self.fc(out)\n",
        "\n",
        "    def forward_repr(self, x):\n",
        "        out = self.conv1(x)\n",
        "        out = self.block1(out)\n",
        "        out = self.block2(out)\n",
        "        out = self.block3(out)\n",
        "        out = self.relu(self.bn1(out))\n",
        "        out = F.avg_pool2d(out, 8)\n",
        "        out = out.view(-1, self.nChannels)\n",
        "        return out"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "O8F8iiZcw_cw",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 136
        },
        "outputId": "258cc732-20c5-4c59-89c6-f87cdf18d212"
      },
      "cell_type": "code",
      "source": [
        "# Load weights\n",
        "!ls\n",
        "model_real_labels   = torch.load('Model_corruptiton_0.0_Acc-tr_99.98.pt')\n",
        "model_random_labels = torch.load('Model_corruptiton_1.0_Acc-tr_99.76.pt')"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "data\r\n",
            "datalab\r\n",
            "Model_corruptiton_0.0_Acc-tr_99.98 (1).pt\r\n",
            "Model_corruptiton_0.0_Acc-tr_99.98.pt\r\n",
            "Model_corruptiton_1.0_Acc-tr_10.24.pt\r\n",
            "Model_corruptiton_1.0_Acc-tr_99.76 (1).pt\r\n",
            "Model_corruptiton_1.0_Acc-tr_99.76.pt\r\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "O2teALSZw_dC",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 612
        },
        "outputId": "d3ff686f-49b7-4442-e658-daa36ea753b1"
      },
      "cell_type": "code",
      "source": [
        "import matplotlib.pyplot as plt\n",
        "from pylab import rcParams\n",
        "def extract_model_parameters_to_array(model):\n",
        "    parameters_array = np.array([])\n",
        "    for name, param in model.cpu().named_parameters():\n",
        "        if param.requires_grad:\n",
        "            flattened_param = param.data.numpy().flatten()\n",
        "            parameters_array = np.append(parameters_array, flattened_param)\n",
        "    return parameters_array\n",
        "\n",
        "# Convert from torch to numpy array\n",
        "parameters_real_labels    =  extract_model_parameters_to_array(model_real_labels)\n",
        "parameters_random_labels  =  extract_model_parameters_to_array(model_random_labels) \n",
        "\n",
        "# Round to 4 decim\n",
        "parameters_real_labels   = parameters_real_labels.round(decimals=4)\n",
        "parameters_random_labels = parameters_random_labels.round(decimals=4)\n",
        "\n",
        "\n",
        "# Plot histogram of the weights\n",
        "rcParams['figure.figsize'] = 30, 10\n",
        "bins = np.linspace(-1, 1, 1000)\n",
        "plt.hist(parameters_real_labels, bins, alpha=0.5, label='Real labels')\n",
        "plt.hist(parameters_random_labels, bins, alpha=0.5, label='Random Labels')\n",
        "plt.legend(loc='upper right')\n",
        "plt.show()\n",
        "\n",
        "#plt.subplot(3,1,1)\n",
        "#plt.hist(parameters_real_labels, bins='auto')\n",
        "#plt.title(\"Histogram- Real labels\")\n",
        "#axes = plt.gca()\n",
        "#axes.set_xlim([-1,1])\n",
        "#plt.subplot(3,1,2)\n",
        "#plt.hist(parameters_gaussain_init, bins='auto')\n",
        "#plt.title(\"Histogram- Random Gaussian Initialization\")\n",
        "#axes = plt.gca()\n",
        "#axes.set_xlim([-1,1])\n",
        "#plt.subplot(3,1,3)\n",
        "#plt.hist(parameters_random_labels, bins='auto')\n",
        "#plt.title(\"Histogram- Random Labels\")\n",
        "#axes = plt.gca()\n",
        "#axes.set_xlim([-1,1])\n",
        "#plt.show()"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAABrsAAAI/CAYAAADHm2LOAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nHAAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7fd7d722d0b8>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "WQbEcH4lw_dK",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        },
        "outputId": "9bbbd8d4-bd4f-45d5-f55e-3ecb90c85f98"
      },
      "cell_type": "code",
      "source": [
        "def entropy(X):\n",
        "    probs = [np.mean(X == c) for c in set(X)]\n",
        "    return np.sum(-p * np.log2(p) for p in probs)\n",
        "\n",
        "# Compute the entorpy for each setss of weights\n",
        "start_time = time.time()\n",
        "entropy_real_labels = entropy(parameters_real_labels)\n",
        "print('Entropy real labels = %f, time=%f [sec]' % (entropy_real_labels, time.time() - start_time ))\n",
        "entropy_random_labels = entropy(parameters_random_labels)\n",
        "print('Entropy random labels = %f, time=%f [sec]' % (entropy_random_labels, time.time() - start_time ))"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Entropy real labels = 12.026414, time=6.057739 [sec]\n",
            "Entropy random labels = 12.943937, time=16.146453 [sec]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "n-yNlt9vw_dY",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Evaluate training acuracy"
      ]
    },
    {
      "metadata": {
        "id": "6Pcr5liOw_da",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# Define general functions\n",
        "\n",
        "class CIFAR10RandomLabels(datasets.CIFAR10):\n",
        "    \"\"\"CIFAR10 dataset, with support for randomly corrupt labels.\n",
        "\n",
        "    Params\n",
        "    ------\n",
        "    corrupt_prob: float\n",
        "        Default 0.0. The probability of a label being replaced with\n",
        "        random label.\n",
        "    num_classes: int\n",
        "        Default 10. The number of classes in the dataset.\n",
        "    \"\"\"\n",
        "    def __init__(self, corrupt_prob=0.0, num_classes=10, **kwargs):\n",
        "        super(CIFAR10RandomLabels, self).__init__(**kwargs)\n",
        "        self.n_classes = num_classes\n",
        "        if corrupt_prob > 0:\n",
        "            self.corrupt_labels(corrupt_prob)\n",
        "\n",
        "    def corrupt_labels(self, corrupt_prob):\n",
        "        labels = np.array(self.train_labels if self.train else self.test_labels)\n",
        "        np.random.seed(12345)\n",
        "        mask = np.random.rand(len(labels)) <= corrupt_prob\n",
        "        rnd_labels = np.random.choice(self.n_classes, mask.sum())\n",
        "        labels[mask] = rnd_labels\n",
        "        # we need to explicitly cast the labels from npy.int64 to\n",
        "        # builtin int type, otherwise pytorch will fail...\n",
        "        labels = [int(x) for x in labels]\n",
        "\n",
        "        if self.train:\n",
        "            self.train_labels = labels\n",
        "        else:\n",
        "            self.test_labels = labels\n",
        "\n",
        "            \n",
        "def get_data_loaders(args, shuffle_train=True):\n",
        "\n",
        "    # Normalization for CIFAR10\n",
        "    normalize = transforms.Normalize(mean=[x/255.0 for x in [125.3, 123.0, 113.9]],\n",
        "                                     std=[x / 255.0 for x in [63.0, 62.1, 66.7]])\n",
        "\n",
        "    if args['data_augmentation']==True:\n",
        "        transform_train = transforms.Compose([\n",
        "          transforms.RandomCrop(32, padding=4),\n",
        "          transforms.RandomHorizontalFlip(),\n",
        "          transforms.ToTensor(),\n",
        "          normalize,\n",
        "          ])\n",
        "    else:\n",
        "        transform_train = transforms.Compose([\n",
        "          transforms.ToTensor(),\n",
        "          normalize,\n",
        "          ])\n",
        "\n",
        "    transform_test = transforms.Compose([\n",
        "        transforms.ToTensor(),\n",
        "        normalize\n",
        "        ])\n",
        "\n",
        "    kwargs = {'num_workers': 1, 'pin_memory': True}\n",
        "    train_loader = torch.utils.data.DataLoader(\n",
        "        CIFAR10RandomLabels(root='./data', train=True, download=True,\n",
        "                            transform=transform_train, num_classes=args['num_classes'],\n",
        "                            corrupt_prob=args['label_corrupt_prob']),\n",
        "        batch_size=args['batch_size'], shuffle=shuffle_train, **kwargs)\n",
        "    val_loader = torch.utils.data.DataLoader(\n",
        "        CIFAR10RandomLabels(root='./data', train=False,\n",
        "                            transform=transform_test, num_classes=args['num_classes'],\n",
        "                            corrupt_prob=args['label_corrupt_prob']),\n",
        "        batch_size=args['batch_size'], shuffle=False, **kwargs)\n",
        "    \n",
        "    # Labels of CIFAR10 dataset\n",
        "    label_translation = ['airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog',\n",
        "                         'horse', 'ship', 'truck']\n",
        "\n",
        "    return train_loader, val_loader, label_translation"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "fN-aHeYxw_di",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "# Define general functions for training\n",
        "def validate_epoch(val_loader, model, criterion, epoch, args):\n",
        "    \"\"\"Perform validation on the validation set\"\"\"\n",
        "    batch_time = AverageMeter()\n",
        "    losses = AverageMeter()\n",
        "    top1 = AverageMeter()\n",
        "\n",
        "    # switch to evaluate mode\n",
        "    model.eval()\n",
        "\n",
        "    for i, (input, target) in enumerate(val_loader):\n",
        "        target = target.cuda(async=True)\n",
        "        input = input.cuda()\n",
        "        input_var = torch.autograd.Variable(input, volatile=True)\n",
        "        target_var = torch.autograd.Variable(target, volatile=True)\n",
        "\n",
        "        # compute output\n",
        "        output = model(input_var)\n",
        "        loss = criterion(output, target_var)\n",
        "\n",
        "        # measure accuracy and record loss\n",
        "        prec1 = accuracy(output.data, target, topk=(1,))[0]\n",
        "        losses.update(loss.data[0], input.size(0))\n",
        "        top1.update(prec1[0], input.size(0))\n",
        "\n",
        "    return losses.avg, top1.avg\n",
        "\n",
        "\n",
        "class AverageMeter(object):\n",
        "    \"\"\"Computes and stores the average and current value\"\"\"\n",
        "    def __init__(self):\n",
        "        self.reset()\n",
        "\n",
        "    def reset(self):\n",
        "        self.val = 0\n",
        "        self.avg = 0\n",
        "        self.sum = 0\n",
        "        self.count = 0\n",
        "\n",
        "    def update(self, val, n=1):\n",
        "        self.val = val\n",
        "        self.sum += val * n\n",
        "        self.count += n\n",
        "        self.avg = self.sum / self.count\n",
        "\n",
        "\n",
        "def accuracy(output, target, topk=(1,)):\n",
        "    \"\"\"Computes the precision@k for the specified values of k\"\"\"\n",
        "    maxk = max(topk)\n",
        "    batch_size = target.size(0)\n",
        "\n",
        "    _, pred = output.topk(maxk, 1, True, True)\n",
        "    pred = pred.t()\n",
        "    correct = pred.eq(target.view(1, -1).expand_as(pred))\n",
        "\n",
        "    res = []\n",
        "    for k in topk:\n",
        "        correct_k = correct[:k].view(-1).float().sum(0)\n",
        "        res.append(correct_k.mul_(100.0 / batch_size))\n",
        "    return res\n",
        "\n",
        "\n",
        "def adding_noise_to_model(model, sigma=1.0):\n",
        "    noisy_model = copy.deepcopy(model)\n",
        "    for param in noisy_model.parameters():\n",
        "        param.data = param.data.add_(np.random.randn()*sigma)\n",
        "    return noisy_model\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "Lt_4Sbavw_dq",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1785
        },
        "outputId": "ddce164a-8207-4740-a282-eb5bce1463b2"
      },
      "cell_type": "code",
      "source": [
        "# Define hyperparameters\n",
        "args = {}\n",
        "args['num_classes'] = 10 \n",
        "args['data_augmentation'] = False \n",
        "args['label_corrupt_prob'] = 0.0\n",
        "args['batch_size'] = 128\n",
        "\n",
        "\n",
        "# Load model\n",
        "labels_state = [True, False]\n",
        "real_labels = False\n",
        "\n",
        "for real_labels in labels_state:\n",
        "    print('Label state: real_labels=%d' % (real_labels) )\n",
        "    if real_labels == True:\n",
        "        args['label_corrupt_prob'] = 0.0 # replace with 1.0 for totaly random labels\n",
        "        model = model_real_labels\n",
        "    else:\n",
        "        args['label_corrupt_prob'] = 1.0 # replace with 1.0 for totaly random labels\n",
        "        model = model_random_labels\n",
        "\n",
        "    model = torch.nn.DataParallel(model).cuda()\n",
        "    model = model.cuda() \n",
        "\n",
        "    # Load the data\n",
        "    train_loader, val_loader, label_translation = get_data_loaders(args, shuffle_train=True)\n",
        "\n",
        "    # define loss function (criterion) and optimizer\n",
        "    criterion = nn.CrossEntropyLoss().cuda()\n",
        "\n",
        "    # Iterate on ephoces\n",
        "    sigma_list = list(np.linspace(0.0001, 0.02, num=50))\n",
        "    #sigma_list = [0.0001, 0.0001, 0.0005, 0.001, 0.005, 0.01, 0.05, 0.1, 0.5, 1]\n",
        "    train_accuracy_list = []\n",
        "    for idx, sigma in enumerate(sigma_list):\n",
        "        model_with_noise = adding_noise_to_model(model, sigma)\n",
        "\n",
        "        epoch_time_start = time.time()\n",
        "        tr_loss, tr_prec1 = validate_epoch(train_loader, model_with_noise, criterion, epoch, args)\n",
        "        epoch_time = time.time() - epoch_time_start\n",
        "\n",
        "        train_accuracy_list.append(tr_prec1)\n",
        "\n",
        "        print('%d/%d: Acc-tr: %6.2f, L-tr: %6.4f, Sigma=%f Time=%10.4f'%(idx, len(sigma_list), tr_prec1, tr_loss, sigma, epoch_time))    \n",
        "\n",
        "    if real_labels == True:\n",
        "        train_accuracy_list_real_label = train_accuracy_list\n",
        "    else:\n",
        "        train_accuracy_list_random_label = train_accuracy_list"
      ],
      "execution_count": 57,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Label state: real_labels=1\n",
            "Files already downloaded and verified\n",
            "0/50: Acc-tr:  99.98, L-tr: 0.0051, Sigma=0.000100 Time=   15.2288\n",
            "1/50: Acc-tr:  99.96, L-tr: 0.0072, Sigma=0.000506 Time=   15.4381\n",
            "2/50: Acc-tr:  99.92, L-tr: 0.0080, Sigma=0.000912 Time=   15.5663\n",
            "3/50: Acc-tr:  99.92, L-tr: 0.0081, Sigma=0.001318 Time=   15.5809\n",
            "4/50: Acc-tr:  99.89, L-tr: 0.0059, Sigma=0.001724 Time=   15.5323\n",
            "5/50: Acc-tr:  98.22, L-tr: 0.0512, Sigma=0.002131 Time=   15.6962\n",
            "6/50: Acc-tr:  98.06, L-tr: 0.0571, Sigma=0.002537 Time=   15.6473\n",
            "7/50: Acc-tr:  99.65, L-tr: 0.0126, Sigma=0.002943 Time=   15.7446\n",
            "8/50: Acc-tr:  89.87, L-tr: 0.3385, Sigma=0.003349 Time=   15.6263\n",
            "9/50: Acc-tr:  96.93, L-tr: 0.0902, Sigma=0.003755 Time=   15.6783\n",
            "10/50: Acc-tr:  97.38, L-tr: 0.0808, Sigma=0.004161 Time=   15.6114\n",
            "11/50: Acc-tr:  77.15, L-tr: 1.7577, Sigma=0.004567 Time=   15.6742\n",
            "12/50: Acc-tr:  77.21, L-tr: 0.6724, Sigma=0.004973 Time=   15.6716\n",
            "13/50: Acc-tr:  82.06, L-tr: 0.5617, Sigma=0.005380 Time=   15.6036\n",
            "14/50: Acc-tr:  22.84, L-tr: 2.8556, Sigma=0.005786 Time=   15.6747\n",
            "15/50: Acc-tr:  68.50, L-tr: 0.8530, Sigma=0.006192 Time=   15.6652\n",
            "16/50: Acc-tr:  63.96, L-tr: 3.2946, Sigma=0.006598 Time=   15.6072\n",
            "17/50: Acc-tr:  14.14, L-tr: 2.4349, Sigma=0.007004 Time=   15.5517\n",
            "18/50: Acc-tr:  86.60, L-tr: 0.4873, Sigma=0.007410 Time=   15.6029\n",
            "19/50: Acc-tr:  54.37, L-tr: 5.3538, Sigma=0.007816 Time=   15.6908\n",
            "20/50: Acc-tr:  74.02, L-tr: 1.4436, Sigma=0.008222 Time=   15.5758\n",
            "21/50: Acc-tr:  71.22, L-tr: 1.0771, Sigma=0.008629 Time=   15.5314\n",
            "22/50: Acc-tr:  14.33, L-tr: 2.7347, Sigma=0.009035 Time=   15.5930\n",
            "23/50: Acc-tr:  48.80, L-tr: 2.5510, Sigma=0.009441 Time=   15.5564\n",
            "24/50: Acc-tr:  48.58, L-tr: 1.9158, Sigma=0.009847 Time=   15.6677\n",
            "25/50: Acc-tr:  12.38, L-tr: 5.0820, Sigma=0.010253 Time=   15.5996\n",
            "26/50: Acc-tr:  18.23, L-tr: 3.3144, Sigma=0.010659 Time=   15.6910\n",
            "27/50: Acc-tr:  24.00, L-tr: 92.6577, Sigma=0.011065 Time=   15.6034\n",
            "28/50: Acc-tr:  10.00, L-tr: 7.2720, Sigma=0.011471 Time=   15.6350\n",
            "29/50: Acc-tr:  15.41, L-tr: 2.5290, Sigma=0.011878 Time=   15.5636\n",
            "30/50: Acc-tr:  15.70, L-tr: 268.1513, Sigma=0.012284 Time=   15.6979\n",
            "31/50: Acc-tr:  19.01, L-tr: 4.8424, Sigma=0.012690 Time=   15.6643\n",
            "32/50: Acc-tr:  11.51, L-tr: 2.2985, Sigma=0.013096 Time=   15.5519\n",
            "33/50: Acc-tr:  16.22, L-tr: 2.2465, Sigma=0.013502 Time=   15.5738\n",
            "34/50: Acc-tr:  15.68, L-tr: 3.0042, Sigma=0.013908 Time=   15.6090\n",
            "35/50: Acc-tr:  15.29, L-tr: 10.5038, Sigma=0.014314 Time=   15.5700\n",
            "36/50: Acc-tr:  10.84, L-tr: 171.5564, Sigma=0.014720 Time=   15.5957\n",
            "37/50: Acc-tr:  10.00, L-tr: 109.8044, Sigma=0.015127 Time=   15.5849\n",
            "38/50: Acc-tr:  12.17, L-tr: 64.3123, Sigma=0.015533 Time=   15.6181\n",
            "39/50: Acc-tr:  10.00, L-tr: 2.4467, Sigma=0.015939 Time=   15.5320\n",
            "40/50: Acc-tr:  10.10, L-tr: 30.7678, Sigma=0.016345 Time=   15.5538\n",
            "41/50: Acc-tr:  10.00, L-tr: 2.3081, Sigma=0.016751 Time=   15.4206\n",
            "42/50: Acc-tr:  22.00, L-tr: 10.8438, Sigma=0.017157 Time=   15.5063\n",
            "43/50: Acc-tr:  10.55, L-tr: 2.3161, Sigma=0.017563 Time=   15.6187\n",
            "44/50: Acc-tr:  12.25, L-tr: 3.9001, Sigma=0.017969 Time=   15.6037\n",
            "45/50: Acc-tr:  10.00, L-tr: 281.5588, Sigma=0.018376 Time=   15.5738\n",
            "46/50: Acc-tr:  12.34, L-tr: 141.7646, Sigma=0.018782 Time=   15.5295\n",
            "47/50: Acc-tr:  10.00, L-tr: 225.5392, Sigma=0.019188 Time=   15.4732\n",
            "48/50: Acc-tr:  10.00, L-tr: 3.8412, Sigma=0.019594 Time=   15.5469\n",
            "49/50: Acc-tr:  10.86, L-tr: 2.6905, Sigma=0.020000 Time=   15.7356\n",
            "Label state: real_labels=0\n",
            "Files already downloaded and verified\n",
            "0/50: Acc-tr:  99.76, L-tr: 0.0436, Sigma=0.000100 Time=   15.5387\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "1/50: Acc-tr:  99.70, L-tr: 0.0419, Sigma=0.000506 Time=   15.5489\n",
            "2/50: Acc-tr:  99.39, L-tr: 0.0624, Sigma=0.000912 Time=   15.5735\n",
            "3/50: Acc-tr:  98.36, L-tr: 0.0994, Sigma=0.001318 Time=   15.6940\n",
            "4/50: Acc-tr:  96.34, L-tr: 0.1287, Sigma=0.001724 Time=   15.6578\n",
            "5/50: Acc-tr:  94.94, L-tr: 0.1486, Sigma=0.002131 Time=   15.6189\n",
            "6/50: Acc-tr:  96.91, L-tr: 0.1039, Sigma=0.002537 Time=   15.6899\n",
            "7/50: Acc-tr:  98.68, L-tr: 0.1109, Sigma=0.002943 Time=   15.6369\n",
            "8/50: Acc-tr:  81.15, L-tr: 0.8492, Sigma=0.003349 Time=   15.6547\n",
            "9/50: Acc-tr:  89.15, L-tr: 0.3082, Sigma=0.003755 Time=   15.6221\n",
            "10/50: Acc-tr:  74.72, L-tr: 0.7807, Sigma=0.004161 Time=   15.5968\n",
            "11/50: Acc-tr:  64.34, L-tr: 1.0934, Sigma=0.004567 Time=   15.6223\n",
            "12/50: Acc-tr:  66.81, L-tr: 0.9385, Sigma=0.004973 Time=   15.6358\n",
            "13/50: Acc-tr:  88.33, L-tr: 0.3394, Sigma=0.005380 Time=   15.6657\n",
            "14/50: Acc-tr:  42.84, L-tr: 4.4025, Sigma=0.005786 Time=   15.6722\n",
            "15/50: Acc-tr:  63.75, L-tr: 1.0923, Sigma=0.006192 Time=   15.5734\n",
            "16/50: Acc-tr:  67.32, L-tr: 1.3856, Sigma=0.006598 Time=   15.5923\n",
            "17/50: Acc-tr:  34.77, L-tr: 26.5041, Sigma=0.007004 Time=   15.6275\n",
            "18/50: Acc-tr:  22.49, L-tr: 2.1308, Sigma=0.007410 Time=   15.5723\n",
            "19/50: Acc-tr:  21.75, L-tr: 3.1668, Sigma=0.007816 Time=   15.5704\n",
            "20/50: Acc-tr:  66.08, L-tr: 1.0344, Sigma=0.008222 Time=   15.5412\n",
            "21/50: Acc-tr:  19.65, L-tr: 4.0430, Sigma=0.008629 Time=   15.5240\n",
            "22/50: Acc-tr:  59.29, L-tr: 1.3800, Sigma=0.009035 Time=   15.5485\n",
            "23/50: Acc-tr:  18.97, L-tr: 35.6622, Sigma=0.009441 Time=   15.5695\n",
            "24/50: Acc-tr:  25.18, L-tr: 13.1773, Sigma=0.009847 Time=   15.6111\n",
            "25/50: Acc-tr:  14.36, L-tr: 50.6327, Sigma=0.010253 Time=   15.6003\n",
            "26/50: Acc-tr:  19.44, L-tr: 2.2153, Sigma=0.010659 Time=   15.4585\n",
            "27/50: Acc-tr:  54.51, L-tr: 2.4586, Sigma=0.011065 Time=   15.5340\n",
            "28/50: Acc-tr:  50.95, L-tr: 2.5513, Sigma=0.011471 Time=   15.5282\n",
            "29/50: Acc-tr:  11.40, L-tr: 53.0208, Sigma=0.011878 Time=   15.6209\n",
            "30/50: Acc-tr:  10.21, L-tr: 66.0075, Sigma=0.012284 Time=   15.5938\n",
            "31/50: Acc-tr:  18.52, L-tr: 5.7060, Sigma=0.012690 Time=   15.6189\n",
            "32/50: Acc-tr:  47.37, L-tr: 5.4718, Sigma=0.013096 Time=   15.5670\n",
            "33/50: Acc-tr:  31.94, L-tr: 2.8265, Sigma=0.013502 Time=   15.5896\n",
            "34/50: Acc-tr:  50.72, L-tr: 2.5057, Sigma=0.013908 Time=   15.5891\n",
            "35/50: Acc-tr:  39.33, L-tr: 1.7262, Sigma=0.014314 Time=   15.6043\n",
            "36/50: Acc-tr:  15.41, L-tr: 67.0812, Sigma=0.014720 Time=   15.5883\n",
            "37/50: Acc-tr:  11.78, L-tr: 2.5315, Sigma=0.015127 Time=   15.5945\n",
            "38/50: Acc-tr:  12.64, L-tr: 6.9082, Sigma=0.015533 Time=   15.6305\n",
            "39/50: Acc-tr:  10.25, L-tr: 2.4161, Sigma=0.015939 Time=   15.6130\n",
            "40/50: Acc-tr:  20.00, L-tr: 2.1603, Sigma=0.016345 Time=   15.5745\n",
            "41/50: Acc-tr:  10.45, L-tr: 123.7778, Sigma=0.016751 Time=   15.6604\n",
            "42/50: Acc-tr:  10.06, L-tr: 2.3560, Sigma=0.017157 Time=   15.6025\n",
            "43/50: Acc-tr:  25.82, L-tr: 10.0180, Sigma=0.017563 Time=   15.5928\n",
            "44/50: Acc-tr:  12.17, L-tr: 122.6932, Sigma=0.017969 Time=   15.6018\n",
            "45/50: Acc-tr:  25.52, L-tr: 10.8662, Sigma=0.018376 Time=   15.5816\n",
            "46/50: Acc-tr:  32.08, L-tr: 2.7659, Sigma=0.018782 Time=   15.4948\n",
            "47/50: Acc-tr:  10.13, L-tr: 2.5441, Sigma=0.019188 Time=   15.4649\n",
            "48/50: Acc-tr:  25.83, L-tr: 4.7675, Sigma=0.019594 Time=   15.6177\n",
            "49/50: Acc-tr:  13.00, L-tr: 2.9832, Sigma=0.020000 Time=   15.5893\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "UMeU7My-65gy",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 334
        },
        "outputId": "4e2af8f1-b8de-418e-82b0-d35254b392a9"
      },
      "cell_type": "code",
      "source": [
        "# red dashes, blue squares and green triangles\n",
        "rcParams['figure.figsize'] = 15, 5\n",
        "plt.plot(sigma_list, train_accuracy_list_real_label, 'ro', label='Real labels')\n",
        "plt.plot(sigma_list, train_accuracy_list_random_label, 'bs', label='Random Labels')\n",
        "plt.xlabel('Gaussin Noise - Sigma')\n",
        "plt.ylabel('Accuracy [%]')\n",
        "plt.legend(loc='upper right')\n",
        "plt.show()"
      ],
      "execution_count": 59,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAA3kAAAE9CAYAAAC2pquGAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7fd76d37ada0>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "H52LuHk-w_d2",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "nBIfLedqw_d6",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "nvNFL0qlw_eA",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}