Variable transformation and importance using OLS/regression

variable_transformation.py

import numpy as np
import pandas as pd
import statsmodels.api as sm

transformations = {'log': np.log,
                   'sqrt': np.sqrt,
                   'sqr': lambda x: np.power(x, 2),
                   'cube': lambda x: np.power(x, 3),
                   'cubert': lambda x: np.power(x, -3),
                   'original': lambda x: x}


def run_transformation(df, columns_to_tranform, transformations, y_col):
    y = df[y_col].values.astype(float)
    all_df = []
    for col in columns_to_tranform:
#         print(col)
        for trans_name, trans in transformations.items():
            X = df[col].apply(trans).replace([np.inf, -np.inf, np.nan], 1e-6).values.reshape(-1, 1)
#             print()
            X = sm.add_constant(X)
#             print()
            lm1 = sm.OLS(y, X).fit()
            s = lm1.summary()
            df_results = pd.DataFrame(s.tables[1].data[1:], columns=s.tables[1].data[0],)
            df_results['r2'] = s.tables[0].data[0][-1]
            df_results['transformation'] = trans_name
            df_results['variable'] = col
            all_df.append(df_results)
    return pd.concat(all_df, axis=0, ignore_index=True)
#             return lm1

results = run_transformation(df_merged, columns_to_tranform, transformations, 'VGO T 95')