AutoFE.py

import pandas as pd, numpy as np
from scipy import stats

def IQR(x, axis):
    return np.quantile(x, 0.75, axis=axis) - np.quantile(x, 0.25, axis=axis)

def sharpe(x, axis):
    return np.mean(x, axis=axis) / np.std(x, axis=axis)

def mode(x, axis):
    return stats.mode(x, axis=axis, keepdims=False).mode

def quantile01(a, axis):
    return np.quantile(a, q=0.1, axis=axis)

def quantile02(a, axis):
    return np.quantile(a, q=0.2, axis=axis)

def quantile08(a, axis):
    return np.quantile(a, q=0.8, axis=axis)

def quantile09(a, axis):
    return np.quantile(a, q=0.9, axis=axis)

base_stats = [np.mean, np.prod, np.std, stats.skew, stats.kurtosis, sharpe]
quantile_stats = [np.min, quantile01, quantile02, np.median, quantile08, quantile09, np.max, IQR, np.ptp, mode]
all_stats = base_stats + quantile_stats

range_count = 23

for df in [train, test]:
    
    #base stats
    for fun in all_stats:
        df[f"Agg_Feature_{fun.__name__}"] = fun(
                (df.loc[:, initial_features].values), axis=1)
        
    # count stats
    for i in np.arange(0, range_count):
        df[f"count_{i}"] = (df[initial_features] == i).sum(axis=1)
        
Features = [f for f in train.columns if f not in targets+ids]