How do I build a class that extends the class of one of its init inputs

Question

I've written a wrapper class around some machine learning models (from sklearn) but I want to expose the underlying model attributes and methods directly for compatibility with other code.

As a toy example...

class LogWrapper(BaseEstimator):

    def __init__(self, model):
        self.model = model

    def fit(self, X, y):
        y_log = np.log(y)
        self.model.fit(X, y_log)

    def predict(self, X):
        y_log_pred = self.model.predict(X)
        y_pred = np.exp(y_log_pred)
        return y_pred

# Then you can do 

my_model = LogWrapper(DecisionTreeRegressor(max_depth=4))

However, I would like to access the model's methods without having to write code of the form

my_model.model.decision_path(X)

I'd prefer to write

my_model.decision_path(X)

The desired behaviour is comparable to having the LogWrapper class inherit from the class of model passed to the init rather than from BaseEstimator.

What you want is not inheritance, but python's data model makes it pretty easy to delegate in a wrapper class. See the accepted answer in the linked duplicate — juanpa.arrivillaga, Jul 08 '19 at 23:07
so in your case, you'd simply want something like `def __getattr__(self, attr): return getattr(self.model, attr)` — juanpa.arrivillaga, Jul 08 '19 at 23:08
...although wrapper classes like this that have if conditions in init are rightly frowned upon — Mr_and_Mrs_D, Jul 08 '19 at 23:09
@Mr_and_Mrs_D what do you mean? which class are you referring to? What condition? — juanpa.arrivillaga, Jul 08 '19 at 23:11
Oh not your answer - this one: https://stackoverflow.com/a/56747128/281545 — Mr_and_Mrs_D, Jul 08 '19 at 23:15
But now I look more carefully the condition is in the factory - ignore my commnet — Mr_and_Mrs_D, Jul 08 '19 at 23:16

How do I build a class that extends the class of one of its __init__ inputs

0 Answers0

How do I build a class that extends the class of one of its init inputs