Sigmoid vs Softmax

Course Q&A TensorFlow Developer Professional Convolutional Neural Networks in TensorFlow
1

Hello,

I was trying to train the model given in the assignment using both softmax and sigmoid activation functions.
The sigmoid function was giving better accuracy compared to softmax.
I had read in the previous course that softmax is ideal for multi-class and sigmoid for multi-label problems.
Since the assignment dealt with multi-class problem I would’ve expected softmax to perform better.
Why is this happening?
What am I missing?

Thanks.

2

Hello,

sigmoid is used for binary classification and softmax is used for multi-classification .
You are dealing with horse vs human or cat vs dog so you have to set activation to sigmoid and class_mode= “binary” .
in multi classification (ex: rock,paper,scissors) you have to set activation = “softmax” and class_mode=‘categorical’.

3

@Sreehari_H1 how many classes do you have in your training data?