from ..explainer import Explainer
import numpy as np

class RandomExplainer(Explainer):
    """ Simply returns random (normally distributed) feature attributions.

    This is only for benchmark comparisons. It supports both fully random attributions and random
    attributions that are constant across all explainations.
    """
    def __init__(self, constant=False):
        self.constant = constant
        self.constant_attributions = None

    def attributions(self, X):
        if self.constant:
            if self.constant_attributions is None:
                self.constant_attributions = np.random.randn(X.shape[1])
            return np.tile(self.constant_attributions, (X.shape[0],1))
        else:
            return np.random.randn(*X.shape)