How can I control the topology of neural network when using MLPClassifier?

Question

I am trying to implement a 3 input logic gate us neural network using MLPClassifier. I want to customize the topology of the neural network so every neuron will have no more than 2 inputs.(by default all inputs are connected to all neurons) For that I was trying to set few components of the weight matrix (.coefs_) to 0 so no connectivity will be created (see code bellow) - but I did not succeed. Any idea how to do it right ?

from sklearn.neural_network import MLPClassifier
X = [[0.,0.,0.], [0.,0.,1.], [0.,1.,0.], [0.,1.,1.],[1.,0.,0.], [1.,0.,1.], [1.,1.,0.], [1.,1.,1.]]
y = [1, 0, 0, 1, 0, 1, 0, 1]
clf = MLPClassifier(alpha=1e-5, random_state=21, activation='logistic', solver='lbfgs',hidden_layer_sizes=(2),tol=0.0001,max_iter=100)
#clf.coefs_[0][2][1]=0   # Not working 
clf.fit(X, y)
clf.coefs_

Answer 1

Scikit-learn is not the right tool for the job. Deep learning frameworks such as Tensorflow or PyTorch allow you to create custom neural network architectures.