mnist with naive binary-tree-like hierarchical softmax

treemnist.py

from __future__ import print_function
import argparse
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torchvision import datasets, transforms
import torchvision

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.num_classes = 10
        # self.conv1 = nn.Conv2d(1, 20, 5, 1)
        # self.conv2 = nn.Conv2d(20, 50, 5, 1)
        # self.fc1 = nn.Linear(4*4*50, 500)
        self.resnet = torchvision.models.__dict__['resnet152'](channels=1)
        self.fc2 = nn.Linear(1000, self.num_classes-1)

    def forward(self, x):
        # x = F.relu(self.conv1(x))
        # x = F.max_pool2d(x, 2, 2)
        # x = F.relu(self.conv2(x))
        # x = F.max_pool2d(x, 2, 2)
        # x = x.view(-1, 4*4*50)
        # x = F.relu(self.fc1(x))
        x = self.resnet(x)
        x = self.fc2(x)
        return F.logsigmoid(x)
        # return F.log_softmax(x, dim=1)

    def traverse_test(self, log_prob, i, accumulate_log_prob):
        assert i >= 0
        assert len(log_prob.shape) == 1
        if i >= self.num_classes-1:
            return accumulate_log_prob, i - self.num_classes + 1
        left_child = 2*i+1
        right_child = 2*i+2
        # import ipdb
        # ipdb.set_trace()
        if log_prob[i].exp() > 0.5:
            lp = log_prob[i]
            child = left_child
        else:
            lp = (1-log_prob[i].exp()).log()
            child = right_child
        next_log_prob = lp + accumulate_log_prob
        # child = left_child if log_prob[i].exp() > 0.5 else right_child
        return self.traverse_test(log_prob, child, next_log_prob)

    def traverse_train(self, log_prob, i, accumulate_log_prob):
        assert i < self.num_classes-1 and i >= 0
        assert len(log_prob.shape) == 1
        if i == 0:
            return accumulate_log_prob + log_prob[0]
        parent = (i-1) // 2
        assert i == 2*parent+1 or i == 2*parent+2
        lp = log_prob[parent] if i == 2*parent+1 else (1-log_prob[parent].exp()).log()
        next_log_prob = lp + accumulate_log_prob
        return self.traverse_train(log_prob, parent, next_log_prob)

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)
        assert(len(target.shape) == 1)
        parent = lambda x: (x-1)//2
        losses = [model.traverse_train(
            output[i, :], parent(t+model.num_classes-1), 0) for i, t in enumerate(target)]
        loss = -sum(losses) / len(losses)
        # loss = F.nll_loss(output, target)
        loss.backward()
        optimizer.step()
        if batch_idx % args.log_interval == 0:
            print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
                epoch, batch_idx * len(data), len(train_loader.dataset),
                100. * batch_idx / len(train_loader), loss.item()))

def test(args, model, device, test_loader):
    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)
            loss, pred = zip(*[model.traverse_test(o, 0, 0) for o in output])
            for l in loss:
                test_loss += l.item()
            # 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 += target.eq(torch.tensor(pred).to(device)).sum().item()
            # correct += pred.eq(target.view_as(pred)).sum().item()

    test_loss /= len(test_loader.dataset)

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

def main():
    # Training settings
    parser = argparse.ArgumentParser(description='PyTorch MNIST Example')
    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=120, metavar='N',
                        help='number of epochs to train (default: 10)')
    parser.add_argument('--lr', type=float, default=0.01, metavar='LR',
                        help='learning rate (default: 0.01)')
    parser.add_argument('--momentum', type=float, default=0.5, metavar='M',
                        help='SGD momentum (default: 0.5)')
    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=10, 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')
    args = parser.parse_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)


    model = Net().to(device)
    optimizer = optim.SGD(model.parameters(), lr=args.lr, momentum=args.momentum)

    for epoch in range(1, args.epochs + 1):
        train(args, model, device, train_loader, optimizer, epoch)
        test(args, model, device, test_loader)

    if (args.save_model):
        torch.save(model.state_dict(),"mnist_cnn.pt")
        
if __name__ == '__main__':
    main()