from math import log

def information_value(feature, target, num_bucket=10):

    data = pd.DataFrame({'tr': target})
    data['bucket'] = pd.qcut(feature, q=num_bucket, duplicates='drop')
    data['cnt'] = 1

    all_tr = data['tr'].mean()
    all_cnt = data['cnt'].sum()

    data_agg = data.groupby('bucket').agg({'tr': 'mean', 'cnt': 'sum'})

    data_agg['tr'] = np.clip(data_agg['tr'], 0.001, 0.999)
    a = data_agg['tr']/all_tr
    b = (1 - data_agg['tr'])/(1-all_tr)
    mult = data_agg['cnt']/all_cnt

    data_agg['iv'] = (a - b)*(np.log(a) - np.log(b)) * mult

    return data_agg['iv'].sum()

information_value(X_train[X_train.columns[5]], pd.Series(y))