Finding probability of a class in multiclass classification

Question

I have made a Tf/Keras CNN Model which is a binary classifier. The model classifies if an image belongs to Class A or Class B. How do I make the model return if an image is a certain percent of Class A and a certain percent of Class B.

This is how I want my outputs.

Kindly help.

Answer 1

Since it's a binary classifier, we have the output layer's activation function as Sigmoid. In sigmoid, the outcome value ranges from 0 to 1. If the outcome is the probability for Class A is x, then the probability for Class B is 1-x.

Therefore the code would be,

pred = model.predict(img)
pred = pred.flatten().flatten()
classA_percent = float(pred)
classB_percent = float(1.0-pred)
pred_class = (pred > 0.5).astype(int)