02. PyTorch Neural Network Classification (timestamp: 12:34:46 - challenge: minimise the loss on test data on CIrcleModelV3)

[tiff78210241911]

Hi, I'm attempting to improve my model to have a higher accuracy than 80% on test data. I have added two more layers (2 -> 5 ->10 -> 10 -> 5 -> 1) so 5 layers in total and changed the learning rate to 0.01. However, neither the loss nor the accuracy is improving. I'm not sure why. Any feedback would be greatly appreciated. Thanks!

torch.manual_seed(42)
torch.cuda.manual_seed(42)

epochs = 1000
# put all data to target devices
X_train, y_train = X_train.to(device), y_train.to(device)
X_test, y_test = X_test.to(device), y_test.to(device)

for epoch in range(epochs):
  ### training
  model_4.train()

  # 1. forward pass
  y_logits = model_4(X_train).squeeze()
  y_preds = torch.round(torch.sigmoid(y_logits))

  # 2. calculate the loss
  loss = loss_fn(y_logits, y_train)
  acc = accurancy_fn(y_true=y_train,
                     y_pred=y_preds)
  
  # 3. optimiser zero grad
  optimsier.zero_grad()

  # 4. loss backward
  loss.backward()

  # 5. optimer step
  optimiser.step()

  ### testing
  model_4.eval()

  with torch.inference_mode():
    # 1. forward pass
    test_logits = model_4(X_test).squeeze()
    test_preds = torch.round(torch.sigmoid(test_logits))

    # 2. calculate the loss
    test_loss = loss_fn(test_logits, y_test)
    test_acc = accurancy_fn(y_true=y_test,
                            y_pred=test_preds)
  
  # print out some stats
  if epoch % 100 == 0:
    print(f"Epoch: {epoch}  | Loss: {loss:.5f}, Acc : {acc:.2f}% | Test loss: {test_loss:.5f}, Test acc: {test_acc:.2f}%")