[問題] numpy 區域合併?

[abc95007 (別理我)]

假設有個矩陣

[ 1, 2, 3, 4, 5, 6]
[ 4, 5, 6, 7, 8, 9]
[ 1, 2, 3, 4, 5, 6]
[ 4, 5, 6, 7, 8, 9]

想要區域性2x2平均回填回去

[3, 3, 5, 5, 7, 7]
[3, 3, 5, 5, 7, 7]
[3, 3, 5, 5, 7, 7]
[3, 3, 5, 5, 7, 7]

或是變成

[3, 5, 7]
[3, 5, 7]

我只想到用迴圈解決
感覺沒發揮到 numpy 的威力
是否有用聰明的用法
謝謝!

import numpy as np
matrix1 = np.arange(0,200,1)[np.newaxis,:]
matrix1 = np.reshape(matrix1,(20,10))
matrix2 = np.zeros((20,10))

for i in range(0,20,2):
    for j in range(0,10,2):
        print(matrix1[i, j])
        temp = [matrix1[i,j],matrix1[i+1,j],matrix1[i,j+1],matrix1[i+1,j+1]]
        average = sum(temp)/len(temp)
        matrix2[i, j] = average
        matrix2[i+1, j] = average
        matrix2[i, j+1] = average
        matrix2[i+1, j+1] = average
print(matrix2)

--
※ 發信站: 批踢踢實業坊(ptt.cc), 來自: 114.45.94.219
※ 文章網址: https://www.ptt.cc/bbs/Python/M.1514390044.A.C6C.html
推 KyotoAnime: 這不就是convolution嗎
推 Blankfein: 就是
→ Blankfein: 直接看tensorflow怎麼做的不是比來板上問快嗎XD
推 Blankfein: tf.nn.Convolution, 找method pool, pooling_type=“AV
→ Blankfein: G”
推 CaptainH: 就做個convolution也要TF?
→ elements: https://stackoverflow.com/q/42463172/2721007 換成 n
→ elements: p.mean
→ abc95007: 感謝樓上提供 這比較簡單XD