def test():
l = 0
correct = 0
for i, (x, label) in enumerate(testloader):
x = Tensor(x.view(-1, INPUT_SIZE).numpy(), requires_grad=False)
label = Tensor(label.numpy().reshape(label.size(0), 1),
requires_grad=False)
x_ = model(x)
x_label = Tensor((x_.shape[0], OUTPUT_SIZE), requires_grad=False)
x_label.put_(label, 1, 1.0)
preds = x_.argmax(1).reshape((label.shape[0], 1))
correct += (preds == label).sum().item()
loss = criterion(x_, x_label, axis=(0, 1))
l += loss
return l / len(testset), correct / len(testset)
def train():
l = 0
correct = 0
for i, (x, label) in enumerate(trainloader):
x = Tensor(x.view(x.size(0), 1, INPUT_SIZE).numpy(), requires_grad=False)
label = Tensor(label.numpy(), requires_grad=False).reshape((x.size(0), 1, 1))
x_ = model(x)
x_label = Tensor((x.size(0), OUTPUT_SIZE, 1), requires_grad=False)
x_label.put_(label, 1, 1.0)
preds = x_.argmax(1).reshape((x.size(0), 1, 1))
correct += (preds == label).sum(axis=(0, 1, 2)).item()
loss = criterion(x_, x_label, axis=(0, 1, 2))
l += loss.item()
loss.backward()
optim.step()
optim.zero_grad()
return l / len(trainset), correct / len(trainset)