Tag: mnist

# MNIST data image of shape 28 * 28 = 784
linear = torch.nn.Linear(784, 10, bias=True).to(device)

# define cost/loss & optimizer
criterion = torch.nn.CrossEntropyLoss().to(device)
optimizer = torch.optim.SGD(linear.parameters(), lr=0.1)

모델은 선형모델을 사용하며 이미지의 크기가 28x28이므로 28*28=784의 차원을 가지는 입력을 받도록 정의한다.

for epoch in range(training_epochs):
    avg_cost = 0
    total_batch = len(data_loader)

    for X, Y in data_loader:
        # reshape input image into [batch_size by 784]
        # label is not one-hot encoded
        X = X.view(-1, 28 * 28).to(device)
        Y = Y.to(device)

        hypothesis = linear(X)
        cost = criterion(hypothesis, Y)
        
        optimizer.zero_grad()
        cost.backward()
        optimizer.step()

        avg_cost += cost / total_batch

    print('Epoch:', '%04d' % (epoch + 1), 'cost =', '{:.9f}'.format(avg_cost))

print('Learning finished')

'''output
Epoch: 0001 cost = 0.535468459
Epoch: 0002 cost = 0.359274179
Epoch: 0003 cost = 0.331187516
Epoch: 0004 cost = 0.316578031
Epoch: 0005 cost = 0.307158142
Epoch: 0006 cost = 0.300180674
Epoch: 0007 cost = 0.295130163
Epoch: 0008 cost = 0.290851504
Epoch: 0009 cost = 0.287417084
Epoch: 0010 cost = 0.284379542
Epoch: 0011 cost = 0.281825215
Epoch: 0012 cost = 0.279800713
Epoch: 0013 cost = 0.277809024
Epoch: 0014 cost = 0.276154280
Epoch: 0015 cost = 0.274440825
Learning finished
'''