0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 1 year has passed since last update.

scipy.stats: 幾何平均と調和平均 gmean, hmean

Last updated at Posted at 2022-05-26

1. 幾何平均 gmean

gmean(a, axis=0, dtype=None, weights=None)

from scipy.stats import gmean
x = [1.2, 3.1, 4.6, 2.2, 5.7]
gmean(x)
2.9263054868319522

gmean 関数を使わずに計算すると以下のようになる。

import numpy as np
np.exp(np.mean(np.log(x)))
2.9263054868319522
np.prod(x)**(1/len(x)) # 非推奨
2.9263054868319527
以下のようなテストデータ平均値=50標準偏差=10)で幾何平均を求めてみる
from scipy.stats import norm
z = norm.rvs(loc=50, scale=10, size=200)
import matplotlib.pyplot as plt
plt.hist(z, bins=30);

output_6_0.png

サンプルサイズ 200 で,教科書的な定義式通りの np.prod(z)**(1/len(z)) は途中でオーバーフローしてしまい inf を返す。

gmean(z), np.exp(np.mean(np.log(z))), np.prod(z)**(1/len(z)), np.prod(z)
(49.19589704885539, 49.19589704885539, inf, inf)

2. 調和平均 hmean

hmean(a, axis=0, dtype=None)

from scipy.stats import hmean
x = [1.2, 3.1, 4.6, 2.2, 5.7]
hmean(x)
2.4958950838709315

hmean 関数を使わずに計算すると以下のようになる。

1/np.mean(1/np.array(x))
2.4958950838709315
`np.array` を使ったのは`x` がリストのとき `1/x` はエラーになるのでnumpy.ndarray に変換するためである

どうせ `numpy` を使うなら逆数を返す `np.reciprocal` を使えば統一性がある
np.reciprocal(np.mean(np.reciprocal(x)))
2.4958950838709315
0
0
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?