Pytorch deep learning video time stamp 7hr 10 min

[c2vishal]

while plotting the list(epoch_count,loss_values,test_loss) in graph daniel was getting error to convert tensor values in numpy arr
when i used inference_mode() for my plot to graph and i dont need to do the conversion is this the right approach?
code:
#plot the loss curves
with torch.inference_mode():
plt.plot(epoch_count,loss_values,label="Train loss")
plt.plot(epoch_count,test_loss_values,label="Test loss")
plt.title("Training and test loss curves")
plt.ylabel("Loss")
plt.xlabel("Epochs")

[Avalianfeng]

When tensor computation is not performed using a GPU, the generated loss_values and test_loss_values are directly processed on the CPU and can be used directly by matplotlib.

Therefore, there's no need to use np.array(torch.tensor(loss_values).numpy()).

You can directly pass in:
loss_value = loss_function(y_pred, y_train)
loss_values.append(loss.item())

However, you must use item() for processing.

When your tensor contains only one element (e.g., a scalar loss), calling .item() will do two things:

Remove gradients and device: Regardless of whether the tensor is on a GPU or CPU, and regardless of whether it has gradients.

Convert the type: Directly convert this standard PyTorch scalar into a pure Python number (usually float or int).

Because Python's native float lists can be perfectly recognized and plotted by Matplotlib, using .item() is extremely efficient.