Add a wrapper to scikit-learn ?

[Paul-Saves]

I want to add a scikit-learn wrapper. It could be a tool located in surrogate.py.

The idea is for one to be able to use "fit(X,y)" or "predict()" instead of set training value(X,y) and then train() and predict() instead of predict_values().

What do you think ?

[relf]

Do you mean adding fit(X, y) and predict() directly to SurrogateModel, something like:

class SurrogateModel(...):
  ...
  def fit(self, X, y):
     self.set_training_values(X, y)
     self.train()

  def predict(x):
     self.predict_values(x)

or something more involved like implementing properly scikit-learn estimators APIs ?

The first one is trivial but can be confusing for newcomers (Should I use train() or fit(), predict() or predict_values()?). Not sure the added-value worths the trouble.
The second one is more complex but will allow to integrate better with the scikit-learn framework. It would mean to create a dedicated
ScikiLearnAdapter class (in utils/sklearn_adapter.py) taking a SMT SurrogateModel at creation time and maintaining the necessary state required by scikit-learn conventions.
I think I would go with a separate ScikiLearnAdapter in utils even just to expose fit() and predict() impemented like above.