How to create a model with multipe output layers?

[guangweizhao]

Hello,

I have 3 different groups of Y labels that share the same X and A for the same dataset, and I wish to do node-level training for all of them together. I tried to create a GNN model with multiple output layers as follows, but this will report errors when I start to train. Is there any better solution to build a model with multiple output layers? Thanks for any suggestions!

X_1 = GCSConv(32, activation='relu')([X_in, A_in])
X_2 = GCSConv(32, activation='relu')([X_1, A_in])
X_3 = GCSConv(32, activation='relu')([X_2, A_in])

output0 = Dense(data.dim0, activation='softmax')(X_3)
output1 = Dense(data.dim1, activation='softmax')(X_3)
output2 = Dense(data.dim2, activation='sigmoid')(X_3)

model.compile(optimizer=opt, loss={'output0': 'categorical_crossentropy', 'output1': 'categorical_crossentropy', 'output2': 'categorical_crossentropy'}, metrics={'output0': 'accuracy', 'output1': 'accuracy', 'output2': 'accuracy'})