pbnn__no_maxpooling.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "3YM3-zGxLa63"
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "import pdb\n",
    "import torch.nn as nn\n",
    "import math\n",
    "from torch.autograd import Variable\n",
    "from torch.autograd import Function\n",
    "import time\n",
    "from torch.distributions.relaxed_bernoulli import RelaxedBernoulli,LogitRelaxedBernoulli\n",
    "import numpy as np\n",
    "\n",
    "\n",
    "def Binarize(tensor,quant_mode='det'):\n",
    "    if quant_mode=='det':\n",
    "      tensor = tensor.sign()\n",
    "      tensor[tensor==0] = 1\n",
    "      return tensor\n",
    "    else:\n",
    "        return tensor.add_(1).div_(2).add_(torch.rand(tensor.size()).add(-0.5)).clamp_(0,1).round().mul_(2).add_(-1)\n",
    "\n",
    "def sample2(mu, log_sigma2):\n",
    "    eps = torch.randn_like(mu)\n",
    "    s = mu + torch.exp(log_sigma2 / 2) * eps\n",
    "    return s\n",
    "  \n",
    "  \n",
    "def sample_gumbel(shape, eps=1e-20):\n",
    "    unif = torch.rand(*shape).cuda()\n",
    "    g = -torch.log(-torch.log(unif + eps))\n",
    "    return g\n",
    "\n",
    "def sample_gumbel_softmax(logits, temperature):\n",
    "    \"\"\"\n",
    "        Input:\n",
    "        logits: Tensor of log probs, shape = BS x k\n",
    "        temperature = scalar\n",
    "        \n",
    "        Output: Tensor of values sampled from Gumbel softmax.\n",
    "                These will tend towards a one-hot representation in the limit of temp -> 0\n",
    "                shape = BS x k\n",
    "    \"\"\"\n",
    "    g = sample_gumbel(logits.shape)\n",
    "    h = (g + logits)/temperature\n",
    "    h_max = h.max(dim=-1, keepdim=True)[0]\n",
    "    h = h - h_max\n",
    "    cache = torch.exp(h)\n",
    "    y = cache / cache.sum(dim=-1, keepdim=True)\n",
    "    return y\n",
    "  \n",
    "def sampling(mu,sig):\n",
    "  x = Normal(mu,sig)\n",
    "\n",
    "  p = 1 - x.cdf(0)\n",
    "\n",
    "  return p\n",
    "\n",
    "\n",
    "\n",
    "import torch.nn._functions as tnnf\n",
    "\n",
    "\n",
    "\n",
    "    \n",
    "class PBinarizeLinear(nn.Linear):\n",
    "\n",
    "    def __init__(self, *kargs, **kwargs):\n",
    "        super(PBinarizeLinear, self).__init__(*kargs, **kwargs)\n",
    "\n",
    "\n",
    "    def forward(self, input):\n",
    "\n",
    "      \n",
    "      \n",
    "        if not hasattr(self.weight,'org'):\n",
    "            self.weight.org=self.weight.data.clone()  \n",
    "            \n",
    "        self.weight.data=Binarize(self.weight.org)\n",
    "\n",
    "        theta = torch.tanh(self.weight)\n",
    "\n",
    "        if input.size(1) != 784:\n",
    "          mu = nn.functional.linear(input,theta)\n",
    "          left = input**2 - (1- input**2)\n",
    "          right = theta**2 - (1-theta**2)\n",
    "          sigma = 1 - nn.functional.linear(left,right)\n",
    "        else:\n",
    "\n",
    "          mu = nn.functional.linear(input,theta)       \n",
    "          sigma = nn.functional.linear(input**2,1-(theta**2))\n",
    "        \n",
    "\n",
    "        m = mu.mean(0,True)\n",
    "        \n",
    "        v = sigma.var(0,True)\n",
    "     \n",
    "        mu = 0.5*(mu-m)/((v+(0.0001)).sqrt()+0.5)\n",
    "        sigma = 0.5**2*sigma/(v+0.0001)\n",
    "\n",
    "        \n",
    "       \n",
    "        out1 = sampling(mu,sigma)\n",
    "\n",
    "        if self.out_features==10:\n",
    "          return mu\n",
    "        else:\n",
    "          return out1\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "class PBinarizeConv2d(nn.Conv2d):\n",
    "\n",
    "    def __init__(self, *kargs, **kwargs):\n",
    "        super(PBinarizeConv2d, self).__init__(*kargs, **kwargs)\n",
    "        \n",
    "\n",
    "    def forward(self, input):\n",
    "      \n",
    "        if not hasattr(self.weight,'org'):\n",
    "            self.weight.org=self.weight.data.clone()  \n",
    "            \n",
    "        self.weight.data=Binarize(self.weight.org)\n",
    "        \n",
    "        theta = torch.tanh(self.weight)\n",
    "        \n",
    "\n",
    "        if input.size(1) != 3:\n",
    "          mu = nn.functional.conv2d(input, theta, None, self.stride,\n",
    "                                   self.padding, self.dilation, self.groups)\n",
    "          left = input**2 - (1- input**2)\n",
    "          right = theta**2 - (1-theta**2)\n",
    "          sigma = 1 - nn.functional.conv2d(left, right, None, self.stride,\n",
    "                                   self.padding, self.dilation, self.groups)\n",
    "        else:\n",
    "          input = torch.tanh(input)\n",
    "          mu = nn.functional.conv2d(input, theta, None, self.stride,\n",
    "                                   self.padding, self.dilation, self.groups)\n",
    "          sigma = nn.functional.conv2d(input**2, 1-(theta**2), None, self.stride,\n",
    "                                   self.padding, self.dilation, self.groups)\n",
    "        \n",
    "#         print(mu.shape)\n",
    "        m = mu.mean((0,2,3),True)\n",
    "        \n",
    "        v = sigma.var((0,2,3)).unsqueeze(0).unsqueeze(-1).unsqueeze(-1)\n",
    "     \n",
    "        mu = 0.5*(mu-m)/((v+(0.0001)).sqrt()+0.5)\n",
    "        sigma = 0.5**2*sigma/(v+0.0001)\n",
    "        \n",
    "        out1 = sampling(mu,sigma)\n",
    "\n",
    "       \n",
    "        \n",
    "        return out1\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 153
    },
    "colab_type": "code",
    "id": "ggHwTvf4MRqC",
    "outputId": "c4a3dbcc-8013-4fe1-eff1-17ffa9961c57"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Net(\n",
      "  (conv1): PBinarizeConv2d(1, 32, kernel_size=(3, 3), stride=(1, 1))\n",
      "  (mp1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  (conv2): PBinarizeConv2d(32, 64, kernel_size=(3, 3), stride=(1, 1))\n",
      "  (mp2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  (fc1): PBinarizeLinear(in_features=36864, out_features=512, bias=True)\n",
      "  (fc2): PBinarizeLinear(in_features=512, out_features=10, bias=True)\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "from __future__ import print_function\n",
    "import argparse\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torch.optim as optim\n",
    "from torchvision import datasets, transforms\n",
    "from torch.autograd import Variable\n",
    "from tqdm import tqdm\n",
    "from torch.distributions.normal import Normal\n",
    "from torch.distributions.relaxed_bernoulli import RelaxedBernoulli\n",
    "from torch.distributions.relaxed_categorical import RelaxedOneHotCategorical\n",
    "\n",
    "from torch.distributions.categorical import Categorical\n",
    "\n",
    "torch.manual_seed(1)\n",
    "\n",
    "train_loader = torch.utils.data.DataLoader(\n",
    "    datasets.MNIST('../data', train=True, download=True,\n",
    "                   transform=transforms.Compose([\n",
    "                       transforms.ToTensor(),\n",
    "                       transforms.Normalize((0.1307,), (0.3081,))\n",
    "                   ])),\n",
    "    batch_size=128, shuffle=True)\n",
    "test_loader = torch.utils.data.DataLoader(\n",
    "    datasets.MNIST('../data', train=False, transform=transforms.Compose([\n",
    "                       transforms.ToTensor(),\n",
    "                       transforms.Normalize((0.1307,), (0.3081,))\n",
    "                   ])),\n",
    "    batch_size=128, shuffle=True)\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "# 32C3 - MP2 - 64C3 - Mp2 - 512FC - SM10c\n",
    "class Net(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(Net, self).__init__()\n",
    "        \n",
    "        self.conv1 = PBinarizeConv2d(1, 32, kernel_size=3)\n",
    "        self.mp1= nn.MaxPool2d(kernel_size=2, stride=2)\n",
    "      \n",
    "        self.conv2 = PBinarizeConv2d(32, 64, kernel_size=3)\n",
    "        self.mp2= nn.MaxPool2d(kernel_size=2, stride=2)\n",
    "     \n",
    "        self.fc1 = PBinarizeLinear(36864, 512)\n",
    "\n",
    "        \n",
    "        self.fc2 = PBinarizeLinear(512, 10)\n",
    "\n",
    "\n",
    "  \n",
    "    def forward(self, x):\n",
    "      \n",
    "#       print(x.shape)\n",
    "        \n",
    "      x = self.conv1(x)\n",
    "      x = self.conv2(x)\n",
    "\n",
    "      x = x.view(x.size(0), -1)\n",
    "#         print(x.size())\n",
    "\n",
    "      x = self.fc1(x)\n",
    "\n",
    "      x = self.fc2(x)\n",
    "\n",
    "\n",
    "      return x\n",
    "  \n",
    "\n",
    "model = Net()\n",
    "\n",
    "print(model)\n",
    "\n",
    "torch.cuda.device('cuda')\n",
    "model.cuda()\n",
    "\n",
    "\n",
    "\n",
    "criterion = nn.CrossEntropyLoss()\n",
    "optimizer = optim.Adam(model.parameters(), lr=0.01)\n",
    "\n",
    "\n",
    "def train(epoch):\n",
    "    model.train()\n",
    "    \n",
    "    losses = []\n",
    "    trainloader = tqdm(train_loader)\n",
    "    \n",
    "    for batch_idx, (data, target) in enumerate(trainloader):\n",
    " \n",
    "        data, target = data.cuda(), target.cuda()\n",
    "        data, target = Variable(data), Variable(target)\n",
    "        optimizer.zero_grad()\n",
    "        output = model(data)\n",
    "\n",
    "        loss = criterion(output, target)\n",
    "\n",
    "\n",
    "\n",
    "        if epoch%40==0 and epoch != 0:\n",
    "            optimizer.param_groups[0]['lr']=optimizer.param_groups[0]['lr']*0.1\n",
    "\n",
    "\n",
    "        \n",
    "        loss.backward()\n",
    "    \n",
    "        for p in list(model.parameters()):\n",
    "            if hasattr(p,'org'):\n",
    "                p.data.copy_(p.org)\n",
    "        optimizer.step()\n",
    "        \n",
    "        for p in list(model.parameters()):\n",
    "            if hasattr(p,'org'):\n",
    "                p.org.copy_(p.data.clamp_(-0.9,0.9))\n",
    "    \n",
    "        losses.append(loss.item())\n",
    "        trainloader.set_postfix(loss=np.mean(losses), epoch=epoch)\n",
    "\n",
    "loss_history = []\n",
    "acc_history = []\n",
    "\n",
    "def test():\n",
    "    model.eval()\n",
    "    test_loss = 0\n",
    "    correct = 0\n",
    "    testloader = tqdm(test_loader)\n",
    "    for data, target in testloader:\n",
    "        data, target = data.cuda(), target.cuda()\n",
    "        with torch.no_grad():\n",
    "          data = Variable(data)\n",
    "        target = Variable(target)\n",
    "        output = model(data)\n",
    "        test_loss += criterion(output, target).item() # sum up batch loss\n",
    "        pred = output.data.max(1, keepdim=True)[1] # get the index of the max log-probability\n",
    "        correct += pred.eq(target.data.view_as(pred)).cpu().sum()\n",
    "        \n",
    "        loss = test_loss / len(test_loader.dataset)\n",
    "        acc = (100. *correct / len(test_loader.dataset)).numpy()\n",
    "\n",
    "        testloader.set_postfix(loss= loss,acc=str(acc)+'%')\n",
    "    \n",
    "    test_loss /= len(test_loader.dataset)\n",
    "    total_acc = correct.item()/len(test_loader.dataset)\n",
    "    loss_history.append(test_loss)\n",
    "    acc_history.append(total_acc)\n",
    "   "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1751
    },
    "colab_type": "code",
    "id": "bxgRGAoMbkCf",
    "outputId": "7ad4524a-ea83-413e-d7ef-67053e0d20c9"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 469/469 [00:37<00:00, 12.69it/s, epoch=0, loss=0.426]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.74it/s, acc=93%, loss=0.00174]\n",
      "100%|██████████| 469/469 [00:38<00:00, 12.33it/s, epoch=1, loss=0.135]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.70it/s, acc=95%, loss=0.00128]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.43it/s, epoch=2, loss=0.133]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.82it/s, acc=96%, loss=0.000873]\n",
      "100%|██████████| 469/469 [00:38<00:00, 12.24it/s, epoch=3, loss=0.115]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.52it/s, acc=95%, loss=0.00109]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.60it/s, epoch=4, loss=0.0959]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.99it/s, acc=96%, loss=0.000876]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.35it/s, epoch=5, loss=0.0714]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.93it/s, acc=96%, loss=0.000863]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.60it/s, epoch=6, loss=0.078]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.90it/s, acc=96%, loss=0.00077]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.38it/s, epoch=7, loss=0.0745]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.86it/s, acc=97%, loss=0.00084]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.60it/s, epoch=8, loss=0.0662]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.63it/s, acc=96%, loss=0.000852]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.35it/s, epoch=9, loss=0.0547]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.82it/s, acc=96%, loss=0.00081]\n",
      "100%|██████████| 469/469 [00:38<00:00, 12.30it/s, epoch=10, loss=0.0594]\n",
      "100%|██████████| 79/79 [00:03<00:00, 23.85it/s, acc=96%, loss=0.000806]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.48it/s, epoch=11, loss=0.072]\n",
      "100%|██████████| 79/79 [00:02<00:00, 28.06it/s, acc=96%, loss=0.00102]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.65it/s, epoch=12, loss=0.055]\n",
      "100%|██████████| 79/79 [00:02<00:00, 26.46it/s, acc=97%, loss=0.000707]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.44it/s, epoch=13, loss=0.0443]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.99it/s, acc=97%, loss=0.00064]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.60it/s, epoch=14, loss=0.0389]\n",
      "100%|██████████| 79/79 [00:03<00:00, 25.96it/s, acc=97%, loss=0.000602]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.43it/s, epoch=15, loss=0.0461]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.93it/s, acc=96%, loss=0.000925]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.60it/s, epoch=16, loss=0.0436]\n",
      "100%|██████████| 79/79 [00:03<00:00, 25.43it/s, acc=97%, loss=0.000662]\n",
      "100%|██████████| 469/469 [00:38<00:00, 12.28it/s, epoch=17, loss=0.0324]\n",
      "100%|██████████| 79/79 [00:03<00:00, 26.20it/s, acc=97%, loss=0.000677]\n",
      "100%|██████████| 469/469 [00:38<00:00, 12.28it/s, epoch=18, loss=0.0361]\n",
      "100%|██████████| 79/79 [00:03<00:00, 24.47it/s, acc=97%, loss=0.000744]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.44it/s, epoch=19, loss=0.0435]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.57it/s, acc=97%, loss=0.000696]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.52it/s, epoch=20, loss=0.0442]\n",
      "100%|██████████| 79/79 [00:03<00:00, 24.43it/s, acc=95%, loss=0.00105]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.36it/s, epoch=21, loss=0.309]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.72it/s, acc=93%, loss=0.00188]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.50it/s, epoch=22, loss=0.109]\n",
      "100%|██████████| 79/79 [00:03<00:00, 24.42it/s, acc=96%, loss=0.000827]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.48it/s, epoch=23, loss=0.0539]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.96it/s, acc=96%, loss=0.00101]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.55it/s, epoch=24, loss=0.0535]\n",
      "100%|██████████| 79/79 [00:03<00:00, 24.42it/s, acc=96%, loss=0.00105]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.41it/s, epoch=25, loss=0.0555]\n",
      "100%|██████████| 79/79 [00:02<00:00, 28.03it/s, acc=97%, loss=0.000743]\n",
      "100%|██████████| 469/469 [00:38<00:00, 12.25it/s, epoch=26, loss=0.0385]\n",
      "100%|██████████| 79/79 [00:03<00:00, 24.37it/s, acc=96%, loss=0.000782]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.46it/s, epoch=27, loss=0.0431]\n",
      "100%|██████████| 79/79 [00:02<00:00, 27.36it/s, acc=97%, loss=0.000747]\n",
      "100%|██████████| 469/469 [00:37<00:00, 12.43it/s, epoch=28, loss=0.0381]\n",
      "100%|██████████| 79/79 [00:03<00:00, 24.42it/s, acc=97%, loss=0.000733]\n",
      "100%|██████████| 469/469 [00:40<00:00, 11.69it/s, epoch=29, loss=0.0336]\n",
      "100%|██████████| 79/79 [00:04<00:00, 16.40it/s, acc=97%, loss=0.00073]\n",
      "100%|██████████| 469/469 [00:54<00:00,  9.75it/s, epoch=30, loss=0.037]\n",
      "100%|██████████| 79/79 [00:04<00:00, 16.54it/s, acc=97%, loss=0.000712]\n",
      "100%|██████████| 469/469 [00:53<00:00,  8.23it/s, epoch=31, loss=0.0438]\n",
      "100%|██████████| 79/79 [00:05<00:00, 14.83it/s, acc=97%, loss=0.000713]\n",
      "100%|██████████| 469/469 [00:53<00:00,  9.18it/s, epoch=32, loss=0.0385]\n",
      "100%|██████████| 79/79 [00:04<00:00, 16.96it/s, acc=97%, loss=0.000631]\n",
      "100%|██████████| 469/469 [00:52<00:00,  8.91it/s, epoch=33, loss=0.0326]\n",
      "100%|██████████| 79/79 [00:04<00:00, 17.21it/s, acc=97%, loss=0.000766]\n",
      "100%|██████████| 469/469 [00:53<00:00,  9.30it/s, epoch=34, loss=0.0382]\n",
      "100%|██████████| 79/79 [00:04<00:00, 17.47it/s, acc=96%, loss=0.000872]\n",
      "100%|██████████| 469/469 [00:52<00:00,  8.90it/s, epoch=35, loss=0.0409]\n",
      "100%|██████████| 79/79 [00:04<00:00, 17.57it/s, acc=97%, loss=0.000635]\n",
      "100%|██████████| 469/469 [00:53<00:00,  8.71it/s, epoch=36, loss=0.0309]\n",
      "100%|██████████| 79/79 [00:04<00:00, 16.96it/s, acc=97%, loss=0.000671]\n",
      "100%|██████████| 469/469 [00:53<00:00,  8.58it/s, epoch=37, loss=0.0293]\n",
      "100%|██████████| 79/79 [00:05<00:00, 14.50it/s, acc=97%, loss=0.000686]\n",
      "100%|██████████| 469/469 [00:55<00:00,  8.99it/s, epoch=38, loss=0.0281]\n",
      "100%|██████████| 79/79 [00:04<00:00, 16.26it/s, acc=96%, loss=0.000834]\n",
      "100%|██████████| 469/469 [00:52<00:00, 10.20it/s, epoch=39, loss=0.0318]\n",
      "100%|██████████| 79/79 [00:04<00:00, 16.97it/s, acc=97%, loss=0.000607]\n",
      "100%|██████████| 469/469 [00:54<00:00,  8.62it/s, epoch=40, loss=0.0262]\n",
      "100%|██████████| 79/79 [00:04<00:00, 17.14it/s, acc=97%, loss=0.000619]\n",
      "100%|██████████| 469/469 [00:53<00:00,  8.81it/s, epoch=41, loss=0.026]\n",
      "100%|██████████| 79/79 [00:04<00:00, 19.60it/s, acc=97%, loss=0.000611]\n",
      "100%|██████████| 469/469 [00:53<00:00,  7.96it/s, epoch=42, loss=0.0261]\n",
      "100%|██████████| 79/79 [00:05<00:00, 15.33it/s, acc=97%, loss=0.000637]\n",
      "100%|██████████| 469/469 [00:54<00:00,  8.68it/s, epoch=43, loss=0.0269]\n",
      "100%|██████████| 79/79 [00:04<00:00, 17.33it/s, acc=97%, loss=0.00063]\n",
      "100%|██████████| 469/469 [00:53<00:00,  8.74it/s, epoch=44, loss=0.0265]\n",
      "100%|██████████| 79/79 [00:04<00:00, 18.21it/s, acc=97%, loss=0.000628]\n",
      "100%|██████████| 469/469 [00:53<00:00,  8.79it/s, epoch=45, loss=0.0261]\n",
      "100%|██████████| 79/79 [00:04<00:00, 16.70it/s, acc=97%, loss=0.000597]\n",
      "100%|██████████| 469/469 [00:53<00:00,  8.39it/s, epoch=46, loss=0.0263]\n",
      "100%|██████████| 79/79 [00:04<00:00, 16.77it/s, acc=97%, loss=0.000636]\n",
      "100%|██████████| 469/469 [00:54<00:00,  8.63it/s, epoch=47, loss=0.0267]\n",
      "100%|██████████| 79/79 [00:04<00:00, 17.64it/s, acc=97%, loss=0.000602]\n",
      "100%|██████████| 469/469 [00:53<00:00,  9.38it/s, epoch=48, loss=0.0269]\n",
      "100%|██████████| 79/79 [00:04<00:00, 18.06it/s, acc=97%, loss=0.000614]\n",
      "100%|██████████| 469/469 [00:52<00:00,  8.08it/s, epoch=49, loss=0.0263]\n",
      "100%|██████████| 79/79 [00:05<00:00, 14.99it/s, acc=97%, loss=0.000617]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 26min 27s, sys: 10min 11s, total: 36min 39s\n",
      "Wall time: 39min 50s\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "%%%time\n",
    "for epoch in range(50):\n",
    "    train(epoch)\n",
    "    test()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 299
    },
    "colab_type": "code",
    "id": "DiDjIjU6Mf_y",
    "outputId": "120427f9-f39d-4be4-e93d-947a74d17e06"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.001736596905440092, 0.001281662294268608, 0.0008732733119279146, 0.0010918957334011793, 0.0008760778043419123, 0.0008630540685728192, 0.0007701293516904115, 0.0008397845501080156, 0.0008522708643227816, 0.0008104524310678244, 0.0008056854465976357, 0.0010161941606551409, 0.0007072486620396376, 0.0006396357720717787, 0.0006024427680298686, 0.0009253880463540554, 0.0006618540566414594, 0.0006772508412599563, 0.000743623273819685, 0.0006957912843674421, 0.00105309232249856, 0.0018843002557754516, 0.0008268962286412716, 0.0010066761691123248, 0.0010452915465459227, 0.000743317361176014, 0.0007817392501980067, 0.0007470115900039672, 0.0007327807482331991, 0.000729856539145112, 0.0007115148741751909, 0.0007134033428505063, 0.0006311646495014429, 0.000766249018162489, 0.0008719614084810018, 0.0006346979960799217, 0.0006713732527568936, 0.0006855957452207804, 0.0008336784705519676, 0.000607497564703226, 0.0006194424599409103, 0.0006109369521960617, 0.0006373898047953844, 0.0006304943887516856, 0.0006278757458552718, 0.0005970316888764501, 0.0006355390802025795, 0.0006020292913541198, 0.0006144934881478548, 0.0006173100559040904]\n",
      "[0.9343, 0.9509, 0.9649, 0.9576, 0.9663, 0.9647, 0.9699, 0.9708, 0.9656, 0.9671, 0.9689, 0.9604, 0.9728, 0.9755, 0.9759, 0.9645, 0.9759, 0.9737, 0.9715, 0.9708, 0.9558, 0.9321, 0.9683, 0.9617, 0.9616, 0.9718, 0.9696, 0.9713, 0.9731, 0.9734, 0.9715, 0.9729, 0.9744, 0.9712, 0.966, 0.9738, 0.975, 0.9745, 0.9695, 0.9764, 0.9747, 0.9748, 0.9734, 0.9762, 0.9756, 0.976, 0.9751, 0.9761, 0.9755, 0.9749]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f73917fd4a8>"
      ]
     },
     "execution_count": 41,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAtoAAADTCAYAAAC/frTSAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 864x216 with 2 Axes>"
      ]
     },
     "metadata": {
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(loss_history)\n",
    "\n",
    "print(acc_history)\n",
    "\n",
    "\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "f = plt.figure(figsize=(12,3))\n",
    "ax = f.add_subplot(121)\n",
    "ax.plot(acc_history,label='PBNet')\n",
    "ax.grid()\n",
    "ax.set_ylabel('Accurancy')\n",
    "ax.set_xlabel('Epochs')\n",
    "ax.legend(loc='lower right')\n",
    "\n",
    "ax2 = f.add_subplot(122)\n",
    "ax2.plot(loss_history,label='PBNet')\n",
    "ax2.grid()\n",
    "ax2.set_ylabel('Loss')\n",
    "ax2.set_xlabel('Epochs')\n",
    "ax2.legend(loc='upper right')"
   ]
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [],
   "name": "pbnn_mnist_v7.ipynb",
   "provenance": [],
   "version": "0.3.2"
  },
  "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}