lenet_original.py

import torch
import torch.nn as nn
from torch.optim import Optimizer
from torch.utils.data import DataLoader
from torchvision import datasets
from torchvision.transforms import transforms

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

'''
Step 1:
'''

# MNIST dataset
train_dataset = datasets.MNIST(root='./mnist_data/',
                               train=True, 
                               transform=transforms.ToTensor(),
                               download=True)

test_dataset = datasets.MNIST(root='./mnist_data/',
                              train=False, 
                              transform=transforms.ToTensor())


'''
Step 2: LeNet5
'''

# Modern LeNet uses this layer for C3
class C3_layer_full(nn.Module):
    def __init__(self):
        super(C3_layer_full, self).__init__()
        self.conv_layer = nn.Conv2d(6, 16, kernel_size=5)

    def forward(self, x):
        return self.conv_layer(x)

# Original LeNet uses this layer for C3
class C3_layer(nn.Module):
    def __init__(self):
        super(C3_layer, self).__init__()
        self.ch_in_3 = [[0, 1, 2],
                        [1, 2, 3],
                        [2, 3, 4],
                        [3, 4, 5],
                        [0, 4, 5],
                        [0, 1, 5]] # filter with 3 subset of input channels
        self.ch_in_4 = [[0, 1, 2, 3],
                        [1, 2, 3, 4],
                        [2, 3, 4, 5],
                        [0, 3, 4, 5],
                        [0, 1, 4, 5],
                        [0, 1, 2, 5],
                        [0, 1, 3, 4],
                        [1, 2, 4, 5],
                        [0, 2, 3, 5]] # filter with 4 subset of input channels
        # put implementation here
        pass

    def forward(self, x):
        # put implementation here
        pass
    
class LeNet(nn.Module) :
    def __init__(self) :
        super(LeNet, self).__init__()
        #padding=2 makes 28x28 image into 32x32
        self.C1_layer = nn.Sequential(
                nn.Conv2d(1, 6, kernel_size=5, padding=2),
                nn.Tanh()
                )
        self.P2_layer = nn.Sequential(
                nn.AvgPool2d(kernel_size=2, stride=2),
                nn.Tanh()
                )
        self.C3_layer = nn.Sequential(
                #C3_layer_full(),
                C3_layer(),
                nn.Tanh()
                )
        self.P4_layer = nn.Sequential(
                nn.AvgPool2d(kernel_size=2, stride=2),
                nn.Tanh()
                )
        self.C5_layer = nn.Sequential(
                nn.Linear(5*5*16, 120),
                nn.Tanh()
                )
        self.F6_layer = nn.Sequential(
                nn.Linear(120, 84),
                nn.Tanh()
                )
        self.F7_layer = nn.Linear(84, 10)
        self.tanh = nn.Tanh()
        
    def forward(self, x) :
        output = self.C1_layer(x)
        output = self.P2_layer(output)
        output = self.C3_layer(output)
        output = self.P4_layer(output)
        output = output.view(-1,5*5*16)
        output = self.C5_layer(output)
        output = self.F6_layer(output)
        output = self.F7_layer(output)
        return output

    
'''
Step 3
'''
model = LeNet().to(device)
loss_function = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(model.parameters(), lr=1e-1)

# print total number of trainable parameters
param_ct = sum([p.numel() for p in model.parameters()])
print(f"Total number of trainable parameters: {param_ct}")

'''
Step 4
'''
train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=100, shuffle=True)

import time
start = time.time()
for epoch in range(10) :
    print("{}th epoch starting.".format(epoch))
    for images, labels in train_loader :
        images, labels = images.to(device), labels.to(device)
        
        optimizer.zero_grad()
        train_loss = loss_function(model(images), labels)
        train_loss.backward()

        optimizer.step()
end = time.time()
print("Time ellapsed in training is: {}".format(end - start))


'''
Step 5
'''
test_loss, correct, total = 0, 0, 0

test_loader = torch.utils.data.DataLoader(dataset=test_dataset, batch_size=100, shuffle=False)

for images, labels in test_loader :
    images, labels = images.to(device), labels.to(device)

    output = model(images)
    test_loss += loss_function(output, labels).item()

    pred = output.max(1, keepdim=True)[1]
    correct += pred.eq(labels.view_as(pred)).sum().item()
    
    total += labels.size(0)
            
print('[Test set] Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)\n'.format(
        test_loss /total, correct, total,
        100. * correct / total))