X, y = trainData.sample(batch_size, i)
X = X.permute(1, 0, 2)
# torch.set_printoptions(profile="full")
classifier.reset()
y_pred = classifier.forward(X)
# print("Y_PRED", y_pred)
loss = criterion.forward(y_pred, y)
# print("CUR", loss.item())
totloss += loss.item()
tot2loss += loss.item()
gradLoss = criterion.backward(y_pred, y)
classifier.backward(gradLoss)
layer.gradWhh = torch.clamp(layer.gradWhh, -5, +5)
layer.gradWxh = torch.clamp(layer.gradWxh, -5, +5)
layer.gradWhy = torch.clamp(layer.gradWhy, -5, +5)
layer.gradBy = torch.clamp(layer.gradBy, -5, +5)
layer.gradBh = torch.clamp(layer.gradBh, -5, +5)
layer.Whh -= alpha * layer.gradWhh
layer.Wxh -= alpha * layer.gradWxh
layer.Why -= alpha * layer.gradWhy
layer.Bh -= alpha * layer.gradBh
layer.By -= alpha * layer.gradBy
label = torch.argmax(y_pred, dim=1)
correct += torch.sum(label == y.long()).item()
count += len(y)
print('Epoch', epoch, 'complete')
