Suggest hyperparameter changes during errors

[jsntsay]

For cases where hyperparameter constraints are violated, it may be possible to suggest changes for the user in the error message. A few study participants mentioned that this would be a nice feature to have.

[hirzel]

Let a = {'h_1': a_1, ..., 'h_n': a_n} be the actual hyperparams that triggered a schema validation excception. And let d = {'h_1': d_1, ..., 'h_n': d_n} the default hyperparams. Assume d is valid and does not trigger a validation exception. Then we could start by trying the n possible ways of changing one of the a_i to a corresponding d_i, and checking whether that validates. If yes, we can return that as a suggested change to the user. Then we can try the n * (n-1) possible changes of two changes a_i, a_j. I would not continue past two, because of computational complexity and because an explanation involving three or more hyperparams would get complicated anyway.

[shinnar]

This is an example error message that is now generated:

Invalid configuration for LogisticRegression(activation='relu', penalty='l1', solver='liblinear', multi_class='multinomial', dual=True) due to argument 'activation' was unexpected.
Some possible fixes include:
- Remove unknown key 'activation' and Set penalty='l2', multi_class='auto'
- Remove unknown key 'activation' and Set dual=False, multi_class='auto'
Schema of arguments and their defaults: {'solver': 'lbfgs', 'penalty': 'l2', 'dual': False, 'C': 1.0, 'tol': 0.0001, 'fit_intercept': True, 'intercept_scaling': 1.0, 'class_weight': None, 'random_state': None, 'max_iter': 100, 'multi_class': 'auto', 'verbose': 0, 'warm_start': False, 'n_jobs': None, 'l1_ratio': None}
Value: {'activation': 'relu', 'penalty': 'l1', 'solver': 'liblinear', 'multi_class': 'multinomial', 'dual': True, 'C': 1.0, 'tol': 0.0001, 'fit_intercept': True, 'intercept_scaling': 1.0, 'class_weight': None, 'random_state': None, 'max_iter': 100, 'verbose': 0, 'warm_start': False, 'n_jobs': None, 'l1_ratio': None}