Convolutional NN - W2 - Assignment 2, question about additional layer?

When doing assignment 2 in week 2, in exercise 2, i just wonder why we should use ‘linear’ activation instead of a ‘sigmoid’ (cause we are trying to predict label with 2 value [‘alpaca’, ‘not alpaca’])

Thank you for spending time read my question.

Further down, you’ll see that the model is using sparse categorical crossentropy and “from_logits = True”.

These two factors (essentially, though not in specific detail) cause TensorFlow to use sigmoid and softmax internally, using a more efficient implementation.

1 Like