def training(net, loader, optimizer):
data_loader, x_features, y_features, x_mean, x_var, y_mean, y_var = train_data()
for epoch in range(epochs):
loss = 0
loss_denormalized = 0
for i, data in enumerate(data_loader):
step = epoch * len(data_loader) + i
inputs, labels = data
prediction = net(inputs)
original_prediction = prediction * y_var + y_mean
original_label = labels * y_var + y_mean
# if epoch == epochs - 1:
# print(original_prediction, original_label)
visual_loss = criterion(original_label, original_prediction)
# writer.add_scalar('Train/doesitwork', visual_loss, step)
loss_denormalized += visual_loss
l = criterion(prediction, labels)
loss += l * labels.size(0)
# backpropagation
optimizer.zero_grad()
l.backward()
optimizer.step()
# writer.add_scalar('Train/weights', net.predict.weight[-1], epoch)
# writer.add_scalar('/Step/Loss', l.item(), step)
# net.log_weights(step)
# writer.add_scalar('Epoch/TrainFc/deno', loss_denormalized, epoch)
# writer.add_scalar('Epoch/TrainFc/Loss', loss, epoch)
print('Epoch: %d | Loss: %.4f ' % (epoch + 1, loss))
return "Finished Training"
loss += l * labels.size(0)
# backpropagation
optimizer.zero_grad()
l.backward()
optimizer.step()
# writer.add_scalar('Train/weights', net.predict.weight[-1], epoch)
# writer.add_scalar('/Step/Loss', l.item(), step)
# net.log_weights(step)
# writer.add_scalar('Epoch/TrainFc/deno', loss_denormalized, epoch)
# writer.add_scalar('Epoch/TrainFc/Loss', loss, epoch)
print('Epoch: %d | Loss: %.4f ' % (epoch + 1, loss))
return "Finished Training"
train_loader, x_features, y_features, x_mean, x_var, y_mean, y_var = train_data()
model = FC(x_features, y_features)
optimizer = optim.Adam(model.parameters(), lr=learning_rate)
# if weights were not saves before run training and save parameters
try:
model.load_state_dict(torch.load('modelWeights/paramsFC.ckpt'))
except FileNotFoundError:
training(model, train_loader, optimizer)
torch.save(model.state_dict(), 'modelWeights/paramsFC.ckpt')
# training(model, train_loader, optimizer)