2020-11-19-224416.py

import time
import numpy as np
import pandas as pd


def np_mode(df):
    """
    与えられたDataFrameの各列から最頻値を返す
    Original: https://twitter.com/nkay/status/1328231713919496194
    """
    arr = df.to_numpy()
    max, min, ncol = arr.max(), arr.min(), arr.shape[1]
    offset = np.arange(ncol) * (max-min+1) - min
    return np.bincount((arr+offset).ravel()).reshape(ncol, -1).argmax(1) + min


nrow = 100
ncol = 400_000
np.random.seed(0)
df = pd.DataFrame(np.random.choice([-1,0,1],(nrow,ncol)))

t1 = time.time()
np_mode(df)
t2 = time.time()

print(f"elapsed time: {t2-t1}")
# elapsed time: 0.2317368984222412

Julia ver. : https://gist.github.com/cac1e8d8e864c0d99148024a5cf15fed