Question regarding discrete model prediction layer activation function and model loss function

[ericparakal]

For the case of the discrete model, specifically the model definition in the file kdd99_model.py; why is the prediction layer activation function sigmoid and not softmax as the KDD99 problem is a multi-class classification problem?

tcav/tcav/tcav_examples/discrete/kdd99_model.py

Line 82 in 218a4cd

pred = tf.keras.layers.Dense(n_labels, activation='sigmoid')(net)

Also, why is the from_logits parameter set to True in the SparseCategoricalCrossentropy loss function, if the prediction layer of the model already has a sigmoid activation function?

tcav/tcav/tcav_examples/discrete/kdd99_model.py

Lines 85 to 87 in 218a4cd

	model_full.compile(loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True), \
	metrics=['accuracy'],
	optimizer='adam')