__find_best_split_for_column.py

def __find_best_split_for_column(self, col):
    x = self.data[col]
    unique_values = x.unique()
    if len(unique_values) == 1: return None, None
    information_gain = None
    split = None
    for val in unique_values:
        left = x <= val
        right = x > val
        left_data = self.data[left]
        right_data = self.data[right]
        left_impurity = self.__calculate_impurity_score(left_data[self.target])
        right_impurity = self.__calculate_impurity_score(right_data[self.target])
        score = self.__calculate_information_gain(left_count = len(left_data),
                                                  left_impurity = left_impurity,
                                                  right_count = len(right_data),
                                                  right_impurity = right_impurity)
        if information_gain is None or score > information_gain: 
            information_gain = score 
            split = val
    return information_gain, split

def __calculate_information_gain(self, left_count, left_impurity, right_count, right_impurity):
    return self.impurity_score - ((left_count/len(self.data)) * left_impurity + \
                                  (right_count/len(self.data)) * right_impurity)