Binarizar variables continuas

El siguiente script transforma las variables continuas en múltiples variables binarias (0,1) que indican a qué cuartil pertenece la observación para cada variable.

NOTA: En algunas situaciones esto puede significar una perdida de información, para detalle sobre esto, ver referencias 1 y 2.


Conceptualmente sería así:

Script:

import pandas as pd
import numpy as np 
from scipy.stats.mstats import mquantiles

def qbinarizer(dataset, not_feature, qt = np.linspace(0,1,11)):
    """ Discretization of continuous features """
    print('binarizing dataset...')
    df_new = pd.DataFrame(index=dataset.index)
    headers =dataset.columns
    headers = [x for x in headers if x not in not_feature]
    for i, col in enumerate(headers):
        if len(dataset[col].unique()) > 1:
            q = mquantiles(dataset[col], qt)
            q = sorted(list(set(q)))
            feature = pd.cut(dataset[col], q, include_lowest = True)
            feature = feature.to_frame().astype(str)
            df = pd.get_dummies(feature)
            df_new = pd.concat([df_new, df], axis = 1)
        if i%20 == 0:
            print(i, 'feature binarized.', str(dataset.shape[1]-i)+' remain')
    print('binarized dataset shahpe: ', df_new.shape)
    return df_new

if __name__ == '__main__':
    path = 'https://www.dropbox.com/s/3g4rz8a0lpvkag4/payment.csv?dl=1'
    data = pd.read_csv(path)
    not_feature = ['fecha', 'user_id']
    data_bin =  qbinarizer(data, not_feature, qt = np.linspace(0,1,5))
    data_bin.to_csv('data_bin.csv')

Referencia
1. https://stats.stackexchange.com/questions/68834/what-is-the-benefit-of-breaking-up-a-continuous-predictor-variable
2. http://biostat.mc.vanderbilt.edu/wiki/Main/CatContinuous