mnist_CNN_compression.py

# Created by moritz (wolter@cs.uni-bonn.de) at 17/12/2019
# based on https://github.com/pytorch/examples/blob/master/mnist/main.py

import argparse
import datetime
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import collections
import pywt
import matplotlib.pyplot as plt
from torchvision import datasets, transforms
from torch.optim.lr_scheduler import StepLR
from src.util import compute_parameter_total
from src.wavelet_learning.wavelet_linear import WaveletLayer
from src.fastfood.fastfood import FastFoodLayer
from torch.utils.tensorboard.writer import SummaryWriter


class Net(nn.Module):
    def __init__(self, compression, wavelet=None, wave_dropout=0.0):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 20, 5, 1)
        self.conv2 = nn.Conv2d(20, 64, 5, 1)
        self.dropout1 = nn.Dropout2d(0.25)
        self.dropout2 = nn.Dropout2d(0.5)
        self.wavelet = wavelet
        self.do_dropout = True
        if compression == 'None':
            self.fc1 = torch.nn.Linear(4*4*64, 500)
            self.fc2 = torch.nn.Linear(500, 10)
        elif compression == 'Fastfood':
            assert wavelet is None
            self.fc1 = FastFoodLayer(1024)
            self.fc2 = torch.nn.Linear(1024, 10)
            self.do_dropout = False
        elif compression == 'Wavelet':
            assert wavelet is not None, 'initial wavelet must be set.'
            self.fc1 = WaveletLayer(init_wavelet=wavelet, scales=6, depth=1024, p_drop=wave_dropout)
            self.fc2 = torch.nn.Linear(1024, 10)
            self.do_dropout = False
        else:
            raise ValueError("Compression type Unknown.")

    def forward(self, x):
        x = self.conv1(x)
        # x = F.relu(x)
        x = F.max_pool2d(x, 2)
        x = self.conv2(x)
        x = F.max_pool2d(x, 2)
        if self.do_dropout:
            x = self.dropout1(x)
        x = torch.flatten(x, 1)
        x = self.fc1(x)
        x = F.relu(x)
        if self.do_dropout:
            x = self.dropout2(x)
        x = self.fc2(x)
        output = F.log_softmax(x, dim=1)
        return output

    def wavelet_loss(self):
        if self.wavelet is None:
            return torch.tensor(0.0), torch.tensor(0.0)
        else:
            acl, _, _ = self.fc1.wavelet.alias_cancellation_loss()
            prl, _, _ = self.fc1.wavelet.perfect_reconstruction_loss()
            return acl, prl


def train(args, model, device, train_loader, optimizer, epoch):
    model.train()
    for batch_idx, (data, target) in enumerate(train_loader):
        data, target = data.to(device), target.to(device)
        optimizer.zero_grad()
        output = model(data)
        nll_loss = F.nll_loss(output, target)
        if args.compression == 'Wavelet':
            acl, prl = model.wavelet_loss()
            wvl = acl + prl
            loss = nll_loss + wvl * args.wave_loss_weight
        else:
            wvl = torch.tensor(0.0)
            loss = nll_loss
        loss.backward()
        optimizer.step()
        if batch_idx % args.log_interval == 0:
            print('Train Epoch: {} [{}/{} ({:.0f}%)], Loss: {:.6f}, wvl-Loss: {:.6f}'.format(
                epoch, batch_idx * len(data), len(train_loader.dataset),
                100. * batch_idx / len(train_loader), nll_loss.item(), wvl.item()))


def test(args, model, device, test_loader, test_writer, epoch):
    model.eval()
    test_loss = 0
    correct = 0
    with torch.no_grad():
        for data, target in test_loader:
            data, target = data.to(device), target.to(device)
            output = model(data)
            test_loss += F.nll_loss(output, target, reduction='sum').item()  # sum up batch loss
            pred = output.argmax(dim=1, keepdim=True)  # get the index of the max log-probability
            correct += pred.eq(target.view_as(pred)).sum().item()

    test_loss /= len(test_loader.dataset)
    acl, prl = model.wavelet_loss()
    wvl_loss = acl + prl

    print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
        test_loss, correct, len(test_loader.dataset),
        100. * correct / len(test_loader.dataset)))

    test_writer.add_scalar('test_correct', correct, epoch)
    test_writer.add_scalar('test_loss', test_loss, epoch)
    test_writer.add_scalar('test_acc', 100. * correct / len(test_loader.dataset), epoch)
    test_writer.add_scalar('wvl_loss', wvl_loss, epoch)
    return wvl_loss,  100. * correct / len(test_loader.dataset)


def main():
    # Training settings
    parser = argparse.ArgumentParser(description='PyTorch Network compression on MNIST')
    parser.add_argument('--batch-size', type=int, default=64, metavar='N',
                        help='input batch size for training (default: 64)')
    parser.add_argument('--test-batch-size', type=int, default=1000, metavar='N',
                        help='input batch size for testing (default: 1000)')
    parser.add_argument('--epochs', type=int, default=14, metavar='N',
                        help='number of epochs to train (default: 14)')
    parser.add_argument('--lr', type=float, default=1.0, metavar='LR',
                        help='learning rate (default: 1.0)')
    parser.add_argument('--gamma', type=float, default=0.7, metavar='M',
                        help='Learning rate step gamma (default: 0.7)')
    parser.add_argument('--no-cuda', action='store_true', default=False,
                        help='disables CUDA training')
    parser.add_argument('--seed', type=int, default=1, metavar='S',
                        help='random seed (default: 1)')
    parser.add_argument('--log-interval', type=int, default=250, metavar='N',
                        help='how many batches to wait before logging training status')
    parser.add_argument('--save-model', action='store_true', default=False,
                        help='For Saving the current Model')
    parser.add_argument('--compression', type=str, default='Wavelet',
                        help='Choose the compression mode, None, Wavelet, Fastfood')
    parser.add_argument('--wave_loss_weight', type=float, default=1.0,
                        help='Weight term of the wavelet loss')
    parser.add_argument('--wave_dropout', type=float, default=0.5,
                        help='Wavelet layer dropout probability.')

    args = parser.parse_args()
    print(args)
    use_cuda = not args.no_cuda and torch.cuda.is_available()

    torch.manual_seed(args.seed)

    device = torch.device("cuda" if use_cuda else "cpu")

    kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {}
    train_loader = torch.utils.data.DataLoader(
        datasets.MNIST('../data', train=True, download=True,
                       transform=transforms.Compose([
                           transforms.ToTensor(),
                           transforms.Normalize((0.1307,), (0.3081,))
                       ])),
        batch_size=args.batch_size, shuffle=True, **kwargs)
    test_loader = torch.utils.data.DataLoader(
        datasets.MNIST('../data', train=False, transform=transforms.Compose([
                           transforms.ToTensor(),
                           transforms.Normalize((0.1307,), (0.3081,))
                       ])),
        batch_size=args.test_batch_size, shuffle=True, **kwargs)

    if args.compression == 'Wavelet':
        CustomWavelet = collections.namedtuple('Wavelet', ['dec_lo', 'dec_hi',
                                                   'rec_lo', 'rec_hi', 'name'])
        # init_wavelet = CustomWavelet(
        #     dec_lo=[0, 0, 0.7071067811865476, 0.7071067811865476, 0, 0],
        #     dec_hi=[0, 0, -0.7071067811865476, 0.7071067811865476, 0, 0],
        #     rec_lo=[0, 0, 0.7071067811865476, 0.7071067811865476, 0, 0],
        #     rec_hi=[0, 0, 0.7071067811865476, -0.7071067811865476, 0, 0],
        #     name='customHaar')

        # random init
        init_wavelet = CustomWavelet(
                    dec_lo=np.random.uniform(-1, 1, size=(6)),
                    dec_hi=np.random.uniform(-1, 1, size=(6)),
                    rec_lo=np.random.uniform(-1, 1, size=(6)),
                    rec_hi=np.random.uniform(-1, 1, size=(6)),
                    name='customHaar')
        # init_wavelet = pywt.Wavelet('db6')
    else:
        init_wavelet = None

    model = Net(compression=args.compression, wavelet=init_wavelet, wave_dropout=args.wave_dropout).to(device)
    print('Parameter total:', compute_parameter_total(model))
    optimizer = optim.Adadelta(model.parameters(), lr=args.lr)
    writer = SummaryWriter()

    scheduler = StepLR(optimizer, step_size=1, gamma=args.gamma)
    test_wvl_lst = []
    test_acc_lst = []
    test_wvl_loss, test_acc = test(args, model, device, test_loader, writer, 0)
    test_wvl_lst.append(test_wvl_loss.item())
    test_acc_lst.append(test_acc)
    for epoch in range(1, args.epochs + 1):
        train(args, model, device, train_loader, optimizer, epoch)
        test_wvl_loss, test_acc = test(args, model, device, test_loader, writer, epoch)
        test_wvl_lst.append(test_wvl_loss.item())
        test_acc_lst.append(test_acc)
        scheduler.step()

    if args.save_model:
        torch.save(model.state_dict(), "mnist_cnn.pt")

    print('Parameter total:', compute_parameter_total(model))

    if args.compression == 'Wavelet':
        plt.plot(model.fc1.wavelet.dec_lo.detach().cpu().numpy(), '-*')
        plt.plot(model.fc1.wavelet.dec_hi.detach().cpu().numpy(), '-*')
        plt.plot(model.fc1.wavelet.rec_lo.detach().cpu().numpy(), '-*')
        plt.plot(model.fc1.wavelet.rec_hi.detach().cpu().numpy(), '-*')
        plt.legend(['H_0', 'H_1', 'F_0', 'F_1'])
        plt.show()
    print('done')


if __name__ == '__main__':
    main()