binarization_utils.py

import numpy as np
import pandas as pd

def convert_continuous_df_to_binary_df(df):
    colnames = df.columns
    n = len(df)
    print("Make sure your first column corresponds to the y label")
    print("Converting continuous features to binary features in the dataframe......")

    percentile_ticks = range(1, 101)

    binarized_dict = {}

    for i in range(0, len(colnames)):
        uni = df[colnames[i]].unique()
        if len(uni) == 2:
            binarized_dict[colnames[i]] = np.asarray(df[colnames[i]], dtype=int)
            continue

        uni.sort()
        if len(uni) >= 100:
            uni = np.percentile(uni, percentile_ticks)
        for j in range(len(uni)-1):
            tmp_feature = np.ones(n, dtype=int)
            tmp_name = colnames[i] + "<=" + str(uni[j])

            zero_indices = df[colnames[i]] > uni[j]
            tmp_feature[zero_indices] = 0

            binarized_dict[tmp_name] = tmp_feature


    binarized_df = pd.DataFrame(binarized_dict)
    print("Finish converting continuous features to binary features......")
    return binarized_df