如何加快3个连续for循环的速度，将数据定位到新的矩阵中

Question

我想从experiment创建新的矩阵experiment2。experiment的维数是13500*12000。我想对每10列求和（但保留行号）以减少维数。你知道吗

第一组是列索引之和0-9, 10-19, 20-29..., 13490-13449

第二组是列索引的和5-14, 15-24, 25-34...,13485-13494

将第一组输入偶数索引，第二组输入experiment2的奇数索引。新矩阵减少到13500*2400

我在这里用3个for循环来表示每一组，速度非常慢。怎么在这里跑得更快？你知道吗

window = 10

experiment2 = np.zeros(shape=(13500,2400))
for eachrow in range(13500):
    for unit in range(1200):
        for even in range(0, (2400), 2):
            # each row, sum column 0-9, 10-19,... and input to experiment2 at even
            experiment2[eachrow, even] = sum(experiment[eachrow,:][(10*unit):((10*unit)+10)])

for eachrow in range(13500):
    for unit in range(1200):
        for odd in range(1, (2400), 2):
            # each row, sum column 5-14, 15-24,... and input to experiment2 at odd
            experiment2[eachrow, odd] = sum(experiment[eachrow,:][((10*unit)+5):((10*unit)+10+5)])

Answer 1

我不是一个裸体专家，但我会这样写：

experiment2 = np.zeros(shape=(13500,2400))
xeven, yeven, xodd, yodd = 0, 10, 5, 15

for eachrow in xrange(13500):
    for unit in xrange(1200):
        sumeven = sum(experiment[eachrow,:][xeven:yeven])
        sumodd = sum(experiment[eachrow,:][xodd:yodd])

        for even in xrange(0, 2400, 2):
            experiment2[eachrow, even] = sumeven
            experiment2[eachrow, even + 1] = sumodd

        xeven += 10
        yeven += 10
        xodd += 10
        yodd += 10

其思想是在同一个循环中填充奇数列和偶数列。另一件事，为了避免计算10*unit+...1200次，我将它们存储在xeven, yeven, xodd, yodd中并递增。 我还使用xrange返回一个迭代器（因为您使用的是python2.7），而不是range返回一个列表。你知道吗