LeNet300_100MNIST.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Copy_of_Untitled1_(10).ipynb",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZVn9PfxdBIz-",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import numpy as np\n",
        "np.random.seed(42)\n",
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline\n",
        "import torch\n",
        "import torch.nn as nn\n",
        "from torch.autograd import Variable\n",
        "import torchvision.datasets as datasets\n",
        "import torchvision.transforms as transforms\n",
        "import argparse\n",
        "import os\n",
        "import torch.nn.functional as F\n",
        "import torch.optim as optim"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jKBfmzL2BZrj",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "trans = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5,), (1.0,))])"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "58orrnUKBfep",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "root = './data'\n",
        "if not os.path.exists(root):\n",
        "    os.mkdir(root)\n",
        "train_set = datasets.MNIST(root=root, train=True, transform=trans, download=True)\n",
        "test_set = datasets.MNIST(root=root, train=False, transform=trans, download=True)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Shbot6zPBfaU",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "batch_size = 60"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YcaIjuvRBfXd",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "train_loader = torch.utils.data.DataLoader(\n",
        "                 dataset=train_set,\n",
        "                 batch_size=batch_size,\n",
        "                 shuffle=True)\n",
        "\n",
        "test_loader = torch.utils.data.DataLoader(\n",
        "                 dataset=test_set,\n",
        "                 batch_size=batch_size,\n",
        "                 shuffle=False)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oe1lUdjBBfOi",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def to_var(x, requires_grad=False, volatile=False):\n",
        "    if torch.cuda.is_available():\n",
        "        x = x.cuda()\n",
        "    return Variable(x, requires_grad=requires_grad, volatile=volatile)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wUk3lFuyBzpb",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def weight_prune(model, pruning_perc):\n",
        "    all_weights = []\n",
        "    for p in model.parameters():\n",
        "        if len(p.data.size()) != 1:\n",
        "            all_weights += list(p.cpu().data.abs().numpy().flatten())\n",
        "             \n",
        "    threshold = np.percentile(np.array(all_weights), pruning_perc)\n",
        "    print(threshold)\n",
        "    \n",
        "    masks = []\n",
        "    for p in model.parameters():\n",
        "        if len(p.data.size()) != 1:\n",
        "            pruned_inds = p.data.abs() > threshold\n",
        "            masks.append(pruned_inds.float())\n",
        "    \n",
        "    return masks"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UtNaAppnuhvR",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def diff_init_weight_prune(model, pruning_perc):\n",
        "    all_weights = []\n",
        "    all_init_weights = []\n",
        "    for p in model.parameters():\n",
        "      if len(p.data.size()) != 1:\n",
        "        all_weights += list(p.cpu().data.abs().numpy().flatten())\n",
        "    \n",
        "    for p in model.init_parameters():\n",
        "      if len(p.data.size()) != 1:\n",
        "        all_init_weights += list(p.cpu().data.abs().numpy().flatten())\n",
        "    \n",
        "    threshold = np.percentile(abs(np.array(all_init_weights)) - abs(np.array(all_weights)), pruning_perc)\n",
        "    \n",
        "    print(threshold)\n",
        "    # generate mask\n",
        "    masks = []\n",
        "    for p, p_init in zip(model.parameters(), model.init_parameters()):\n",
        "      if len(p.data.size()) != 1 and len(p_init.data.size()) != 1:\n",
        "        pruned_inds = p_init.data.abs().cuda() - p.data.abs() > threshold\n",
        "        masks.append(pruned_inds.float())\n",
        "    \n",
        "    return masks"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8QGg4HYZB400",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class MaskedLinear(nn.Linear):\n",
        "    def __init__(self, in_features, out_features, bias=True):\n",
        "        super(MaskedLinear, self).__init__(in_features, out_features, bias)\n",
        "        self.mask_flag = False\n",
        "    \n",
        "    def set_mask(self, mask):\n",
        "        self.mask = mask\n",
        "        mask_var = self.get_mask()\n",
        "        self.weight.data = self.weight.data * mask_var.data\n",
        "        self.mask_flag = True\n",
        "    \n",
        "    def get_mask(self):\n",
        "        return to_var(self.mask, requires_grad=False)\n",
        "    \n",
        "    def forward(self, x):\n",
        "        if self.mask_flag == True:\n",
        "            mask_var = self.get_mask()\n",
        "            weight = self.weight * mask_var\n",
        "            return F.linear(x, weight, self.bias)\n",
        "        else:\n",
        "            return F.linear(x, self.weight, self.bias)\n",
        " "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "v4colmQRB4wX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class Lenet_300_100(nn.Module):\n",
        "  def __init__(self):\n",
        "    super(Lenet_300_100, self).__init__()\n",
        "    \n",
        "    self.fc1 = MaskedLinear(28 * 28, 300)\n",
        "    self.relu1 = nn.ReLU(inplace = True)\n",
        "    \n",
        "    self.fc2 = MaskedLinear(300, 100)\n",
        "    self.relu2 = nn.ReLU(inplace = True)\n",
        "    \n",
        "    self.fc3 = MaskedLinear(100, 10)\n",
        "    \n",
        "    self.init_params = self.state_dict()\n",
        "    \n",
        "  def forward(self, x):\n",
        "    x = x.view(x.size(0), 28 * 28)\n",
        "    out = self.fc1(x)\n",
        "    out = self.relu1(out)\n",
        "    out = self.fc2(out)\n",
        "    out = self.relu2(out)\n",
        "    logits = self.fc3(out)\n",
        "    \n",
        "    return logits \n",
        "  \n",
        "  def set_masks(self, masks):\n",
        "    self.fc1.set_mask(masks[0])\n",
        "    self.fc2.set_mask(masks[1])\n",
        "    self.fc3.set_mask(masks[2])\n",
        "    \n",
        "  def reset_parameters(self):\n",
        "    self.load_state_dict(self.init_params)\n",
        "    \n",
        "  def init_parameters(self):\n",
        "    for name in self.init_params.keys():\n",
        "      yield self.init_params[name]\n",
        "        "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6PnWfk1REb7K",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def compute_accuracy(model, loader):\n",
        "\n",
        "    model.eval()\n",
        "\n",
        "    num_correct, num_samples = 0, len(loader.dataset)\n",
        "    for x, y in loader:\n",
        "        x_var = to_var(x, volatile=True)\n",
        "        scores = model(x_var)\n",
        "        _, preds = scores.data.cpu().max(1)\n",
        "        num_correct += (preds == y).sum()\n",
        "\n",
        "    acc = float(num_correct) / num_samples\n",
        "\n",
        "    return acc"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9NwiNUIfvuES",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#print(\"--- {}% parameters pruned ---\".format(prune_rate(model)))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ATxSQn5qEh57",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def train(model, loss_fn, optimizer, epochs, purge_epochs, purge_rate,\n",
        "          loader_train, loader_val=None):\n",
        "\n",
        "    model.train()\n",
        "    \n",
        "    loss_list = []\n",
        "    accuracy_list = []\n",
        "    \n",
        "    for epoch in range(epochs):\n",
        "        if ((epoch + 1) % purge_epochs == 0):\n",
        "          if purge_rate != 0: \n",
        "            print(\"Purging...\")     \n",
        "            masks = diff_init_weight_prune(model, (1 - (1 - purge_rate / 100) ** ((epoch + 1) / purge_epochs)) * 100)\n",
        "            \n",
        "            model.reset_parameters()\n",
        "            model.set_masks(masks)\n",
        "            \n",
        "            prune_rate(model)\n",
        "            \n",
        "        print('Starting epoch %d / %d' % (epoch + 1, epochs))\n",
        "        av_loss = 0\n",
        "        for t, (x, y) in enumerate(loader_train):\n",
        "            x_var, y_var = to_var(x), to_var(y.long())\n",
        "\n",
        "            scores = model(x_var)\n",
        "            loss = loss_fn(scores, y_var)\n",
        "            \n",
        "            optimizer.zero_grad()\n",
        "            loss.backward()\n",
        "            \n",
        "            optimizer.step()\n",
        "            av_loss += loss.item()\n",
        "        \n",
        "        \n",
        "        loss_list.append(av_loss / t)\n",
        "        accuracy_list.append(compute_accuracy(model, test_loader))\n",
        "          \n",
        "    return loss_list, accuracy_list, purge_rate     "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XRv_i2pSI6OI",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def prune_rate(model, verbose=True):\n",
        "    total_nb_param = 0\n",
        "    nb_zero_param = 0\n",
        "\n",
        "    layer_id = 0\n",
        "\n",
        "    for parameter in model.parameters():\n",
        "      \n",
        "        param_this_layer = 1\n",
        "        for dim in parameter.data.size():\n",
        "            param_this_layer *= dim\n",
        "        total_nb_param += param_this_layer\n",
        "\n",
        "        # only pruning linear and conv layers\n",
        "        if len(parameter.data.size()) != 1:\n",
        "            layer_id += 1\n",
        "            zero_param_this_layer = \\\n",
        "                np.count_nonzero(parameter.cpu().data.numpy()==0)\n",
        "            nb_zero_param += zero_param_this_layer\n",
        "\n",
        "            if verbose:\n",
        "                print(\"Layer {} | {} layer | {:.2f}% parameters pruned\" \\\n",
        "                    .format(\n",
        "                        layer_id,\n",
        "                        'Conv' if len(parameter.data.size()) == 4 \\\n",
        "                            else 'Linear',\n",
        "                        100.*zero_param_this_layer/param_this_layer,\n",
        "                        ))\n",
        "                \n",
        "    pruning_perc = 100. * nb_zero_param/total_nb_param\n",
        "    \n",
        "    if verbose:\n",
        "        print(\"Current pruning rate: {:.2f}%\".format(pruning_perc))\n",
        "    \n",
        "    return pruning_perc"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Qn3DlZEtBOb8",
        "colab_type": "code",
        "outputId": "d539142a-b1d2-438a-dc6f-4745487e1bd4",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 866
        }
      },
      "source": [
        "net = Lenet_300_100()\n",
        "\n",
        "if torch.cuda.is_available():\n",
        "    print('CUDA enabled.')\n",
        "    net.cuda()\n",
        "\n",
        "print(\"--- Before retraining ---\")\n",
        "print(\"Accuracy: {}\".format(compute_accuracy(net, test_loader)))\n",
        "\n",
        "# Retrain\n",
        "criterion = nn.CrossEntropyLoss()\n",
        "optimizer = torch.optim.RMSprop(net.parameters(), lr=1.2e-3, \n",
        "                                weight_decay=0.0001)\n",
        "\n",
        "loss, accuracy, _ = train(net, criterion, optimizer, epochs=30, \n",
        "                       purge_epochs=16, purge_rate=75, \n",
        "                       loader_train=train_loader)\n",
        "\n",
        "print(\"--- After retraining ---\")\n",
        "print(\"Accuracy: {}\".format(compute_accuracy(net, test_loader)))\n",
        "prune_rate(net)"
      ],
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "CUDA enabled.\n",
            "--- Before retraining ---\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:4: UserWarning: volatile was removed and now has no effect. Use `with torch.no_grad():` instead.\n",
            "  after removing the cwd from sys.path.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Accuracy: 0.0811\n",
            "Starting epoch 1 / 30\n",
            "Starting epoch 2 / 30\n",
            "Starting epoch 3 / 30\n",
            "Starting epoch 4 / 30\n",
            "Starting epoch 5 / 30\n",
            "Starting epoch 6 / 30\n",
            "Starting epoch 7 / 30\n",
            "Starting epoch 8 / 30\n",
            "Starting epoch 9 / 30\n",
            "Starting epoch 10 / 30\n",
            "Starting epoch 11 / 30\n",
            "Starting epoch 12 / 30\n",
            "Starting epoch 13 / 30\n",
            "Starting epoch 14 / 30\n",
            "Starting epoch 15 / 30\n",
            "Purging...\n",
            "0.02431097999215126\n",
            "Layer 1 | Linear layer | 77.41% parameters pruned\n",
            "Layer 2 | Linear layer | 55.91% parameters pruned\n",
            "Layer 3 | Linear layer | 81.40% parameters pruned\n",
            "Current pruning rate: 74.88%\n",
            "Starting epoch 16 / 30\n",
            "Starting epoch 17 / 30\n",
            "Starting epoch 18 / 30\n",
            "Starting epoch 19 / 30\n",
            "Starting epoch 20 / 30\n",
            "Starting epoch 21 / 30\n",
            "Starting epoch 22 / 30\n",
            "Starting epoch 23 / 30\n",
            "Starting epoch 24 / 30\n",
            "Starting epoch 25 / 30\n",
            "Starting epoch 26 / 30\n",
            "Starting epoch 27 / 30\n",
            "Starting epoch 28 / 30\n",
            "Starting epoch 29 / 30\n",
            "Starting epoch 30 / 30\n",
            "--- After retraining ---\n",
            "Accuracy: 0.9696\n",
            "Layer 1 | Linear layer | 77.41% parameters pruned\n",
            "Layer 2 | Linear layer | 55.91% parameters pruned\n",
            "Layer 3 | Linear layer | 81.40% parameters pruned\n",
            "Current pruning rate: 74.89%\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "74.88541315029444"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 22
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1c59V3qqXp19",
        "colab_type": "code",
        "outputId": "da07ec41-d6ac-46a0-b8ed-4ef34cef583b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 286
        }
      },
      "source": [
        "plt.plot(accuracy)"
      ],
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[<matplotlib.lines.Line2D at 0x7fb8ae0bfdd8>]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 23
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX0AAAD8CAYAAACb4nSYAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ntRkAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tKnR7KAfbXLp",
        "colab_type": "code",
        "outputId": "d3d376e8-d57c-49e6-81ab-918ed3bb38c9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 55
        }
      },
      "source": [
        "print(loss)"
      ],
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[0.37454348518415254, 0.16619072936557316, 0.12167411898397976, 0.09812249641052566, 0.0829204150696506, 0.07506143394356558, 0.06649163151935984, 0.06243602092167152, 0.05639147490180947, 0.052707788548562805, 0.0492002956410828, 0.04684338900209353, 0.044660410952630326, 0.04254783810516288, 0.04237021210605047, 0.47521380722910556, 0.24057531851652386, 0.1678246728203318, 0.1320295231817691, 0.11073972986408838, 0.09726024542546442, 0.08685881405224179, 0.0794089839575128, 0.07227349779135114, 0.06726165280213116, 0.06208377670602979, 0.05898759173319299, 0.05458086603463304, 0.052406776522111904, 0.051122773362247194]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7S0SsUbLNuKM",
        "colab_type": "code",
        "outputId": "8bce3608-d0ae-46ea-f7ee-96900c3c8d18",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 55
        }
      },
      "source": [
        "print(accuracy)"
      ],
      "execution_count": 21,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[0.9409, 0.9643, 0.9577, 0.965, 0.9595, 0.9747, 0.9733, 0.9757, 0.9754, 0.9735, 0.9752, 0.9749, 0.9533, 0.9764, 0.9777, 0.9109, 0.9463, 0.9537, 0.9639, 0.9635, 0.9459, 0.9713, 0.974, 0.9634, 0.9755, 0.9743, 0.9713, 0.9746, 0.9737, 0.9759]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EJ6bqRsK9wbE",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from matplotlib.colors import ListedColormap\n",
        "from matplotlib import colors\n",
        "\n",
        "def visualize(weights, title, figsize=None):\n",
        "  plt.figure(figsize=figsize)\n",
        "  \n",
        "  cmap = ListedColormap(['k', 'w', 'r'])\n",
        "  #cax = ax.matshow(x,cmap=cmap)\n",
        "  data = weights.data.cpu().detach().abs().numpy().flatten()\n",
        "  half = int(np.sqrt(data.size)) + 1\n",
        "  data = np.concatenate((data, np.zeros(half * half - data.size)), axis=0)\n",
        "  data = data.reshape(half, half)\n",
        "  \n",
        "  bounds=[0,0.000000000000001,0.5,1.0]\n",
        "  norm = colors.BoundaryNorm(bounds, cmap.N)\n",
        "  \n",
        "  plt.imshow(data, cmap=cmap, norm=norm)\n",
        "  plt.colorbar()\n",
        "  plt.title(title)\n",
        "  plt.axis('off')\n",
        "  plt.tight_layout()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "uA_bXEVo91rL",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 827
        },
        "outputId": "2f71275d-6073-4a5a-886e-531640ee824d"
      },
      "source": [
        "visualize(net.fc2.weight,'FC weight', (8, 8))"
      ],
      "execution_count": 186,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAu0AAAMWCAYAAABFnLgJAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAVVlqU7sKidaAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 800x800 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Xn74fUpQNvNG",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 286
        },
        "outputId": "08e3f46c-deb3-4ff4-ab3c-247b71c06bee"
      },
      "source": [
        "activations = {}\n",
        "def get_activation(name):\n",
        "    def hook(model, input, output):\n",
        "        activations[name] = output.detach()\n",
        "    \n",
        "    return hook\n",
        "\n",
        "weights = net.fc2.weight.data.cpu().numpy()\n",
        "net.fc2.register_forward_hook(get_activation('fc3'))\n",
        "\n",
        "x = torch.randn(1, 784)\n",
        "x = x.cuda()\n",
        "output = net(x)\n",
        "\n",
        "\n",
        "plt.imshow(weights, interpolation='nearest', aspect='auto')"
      ],
      "execution_count": 47,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7fb8ae5fe4a8>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 47
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAD8CAYAAABn919SAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nEXhAZxiG6RF4QGcYhukR/n+MfEpyvs3YUAAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NspBGnCqtADX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "-"
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}