Using while loop seems a good way to fix convergence stop.

Example of usage

In how to use whle loop seems to be a good start.

General

• Add stop by convergence of gradient.
• Find out how to measure the number of times we call the gradient.
• Separate hyperparameters from normal parameters.
• Add the regularization term to the cost on train, both RBF and MLP.
• Convergence thing.

Non priority

• Add coments in every function
• Possibly add doc.
• Think about better way to define parameters.
• Create functions separating utilities.
• Add axis labels for all plots.
• Generate markdown table.

For nice

• Modify generation_train_test to retrieve test_y with proper size

Select dinamically the limits of the plot.