two_sigma_financial_modelling.py 文件源码

def feature_scaling(self, df):
        df = df.copy()
        # Standardization (centering and scaling) of dataset that removes mean and scales to unit variance
        standard_scaler = StandardScaler()
        numerical_feature_names_of_non_modified_df = TwoSigmaFinModTools._numerical_feature_names
        if any(tuple(df.columns == 'y')):
            if not TwoSigmaFinModTools._is_one_hot_encoder:
                numerical_feature_names_of_non_modified_df = np.concatenate(
                    [TwoSigmaFinModTools._feature_names_num.values, numerical_feature_names_of_non_modified_df.values])
            # Include scaling of y
            y = df['y'].values
            relevant_features = df[numerical_feature_names_of_non_modified_df].columns[
                (df[numerical_feature_names_of_non_modified_df].columns != 'y')
                & (df[numerical_feature_names_of_non_modified_df].columns != 'id')]
            mask = ~df[relevant_features].isnull()
            res = standard_scaler.fit_transform(X=df[relevant_features][mask].values, y=y)
            if (~mask).sum().sum() > 0:
                df = self.standardize_relevant_features(df, relevant_features, res)
            else:
                df.loc[:, tuple(relevant_features)] = res
        else:
            if not TwoSigmaFinModTools._is_one_hot_encoder:
                numerical_feature_names_of_non_modified_df = np.concatenate(
                    [TwoSigmaFinModTools._feature_names_num.values, numerical_feature_names_of_non_modified_df.values])
            relevant_features = df[numerical_feature_names_of_non_modified_df].columns[
                (df[numerical_feature_names_of_non_modified_df].columns != 'id')]
            mask = ~df[relevant_features].isnull()
            res = standard_scaler.fit_transform(df[relevant_features][mask].values)
            if mask.sum().sum() > 0:
                df = self.standardize_relevant_features(df, relevant_features, res)
            else:
                df.loc[:, tuple(relevant_features)] = res
        return df