According to the documentation, it says the 'activation' argument specifies:
"Activation function for the hidden layer"
Does that mean that you cannot use a different activation function in different layers?
One can see from the code (look at uses of
self.activation) that the same function is used for all the hidden layers.
You might want to consider the comments to this question for alternative approaches, generally being a move away from
sklearn and towards a deep learning framework.