带有大量参数的 sklearn 的自定义估计器

Question

I am trying to make custom estimator for sklearn with a lot of params. As i understand, i should do something like this:

    class MyEstim(BaseEstimator):

     def __init__(self, param1, param2):
      super().__init__()

     self.param1 = param1
     self.param2 = param2
     ...

But i am not sure how to do it if i have tens or hundreds of params. It is not a problem to build params - valid values list, but i dont understand how it will be better to pass this list to estimator in order to make it working in sklearn.

Something like this not working for me:

         def __init__(self, **kwargs):
          super().__init__()
          # init params from kwargs

Because, as i understand, there are some conventions in sklearn code, so it can only initialize params which are explicitly defined in __init___ function.

Answer 1

You can use **kwargs as long as your get_params and set_params methods work as expected.

Example:

(taken from a blog post I authored)

class LoggingEstimator(BaseEstimator):
    def __init__(self, est_class=LinearRegression, **kwargs):

        self.est_class = est_class

        # kwargs depend on the model used, so assign them whatever they are
        for key, value in kwargs.items():
            setattr(self, key, value)

        self._param_names = ['est_class'] + list(kwargs.keys())

    def get_params(self, deep=True):
        # Note: we are ignoring the deep parameter
        # this will not work with estimators that have sub-estimators
        # see https://scikit-learn.org/stable/developers/develop.html#get-params-and-set-params
        return {param: getattr(self, param)
                for param in self._param_names}

    def set_params(self, **parameters):
        for parameter, value in parameters.items():
            setattr(self, parameter, value)

        return self