sklearn MLPClassifier - zero hidden layers i e logistic regression

Question

We know that a feed forward neural network with 0 hidden layers (i.e. just an input layer and an output layer) with a sigmoid activation function at the end should be equivalent to logistic regression.

I wish to prove this to be true, but I need to fit 0 hidden layers using the sklearn MLPClassifier module specifically.

My attempt:

my_nn = MLPClassifier(hidden_layer_sizes=(0), alpha = 0, 
                        max_iter=10000)

but this results in an error message:

hidden_layer_sizes must be > 0, got [0, 0].

Is there any way to achieve this using this specific module?

Answer 1

You could try something like this.

my_nn = MLPClassifier(hidden_layer_sizes=(), alpha = 0, max_iter=10000)

This might solve the error you are getting.