Exercise 2 - alpaca_model (linear)

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

In Programming Assignment: Transfer Learning with MobileNet at Exercise 2 - alpaca_model
when I use sigmoid as activation function does not pass test when I use linear as activation function it is pass why?
I think in binary classification should use sigmoid.

2

Please pay attention to the logits flag in the loss function:

loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),

When logits is True, softmax is not required as the output activation function. See this as well.

1 Like
3

Right: the point is that the sigmoid (or softmax) is included, but it’s being handled internally by the loss function instead of explicitly in your code. Here’s a thread which explains why they do it that way.

1 Like