python - 值错误: Chunks and shape must be of the same length/dimension

Question

我读了《数据科学简介。大数据、机器学习等，使用 Python 工具》一书 Chapter4中有一段关于分块矩阵计算的代码:

import dask.array as da
import bcolz as bc
import numpy as np
import dask

n = 1e4 #A

ar = bc.carray(np.arange(n).reshape(n/2,2)  , dtype='float64', rootdir     = 'ar.bcolz', mode = 'w') #B
y  = bc.carray(np.arange(n/2), dtype='float64', rootdir = 'yy.bcolz', mode = 'w') #B,

dax = da.from_array(ar, chunks=(5,5)) #C
dy = da.from_array(y,chunks=(5,5)) #C

XTX = dax.T.dot(dax) #D
Xy  = dax.T.dot(dy) #E

coefficients = np.linalg.inv(XTX.compute()).dot(Xy.compute()) #F

coef = da.from_array(coefficients,chunks=(5,5)) #G

ar.flush() #H
y.flush() #H

predictions = dax.dot(coef).compute() #I
print (predictions)

我得到值错误:

ValueError                                Traceback (most recent call last)
<ipython-input-4-7ae8e9cf2346> in <module>()
     10 
     11 dax = da.from_array(ar, chunks=(5,5)) #C
---> 12 dy = da.from_array(y,chunks=(5,5)) #C
     13 
     14 XTX = dax.T.dot(dax) #D

C:\Users\F\Anaconda3\lib\site-packages\dask\array\core.py in from_array(x, chunks, name, lock, fancy, getitem)
   1868     >>> a = da.from_array(x, chunks=(1000, 1000), lock=True)  # doctest: +SKIP
   1869     """
-> 1870     chunks = normalize_chunks(chunks, x.shape)
   1871     if len(chunks) != len(x.shape):
   1872         raise ValueError("Input array has %d dimensions but the supplied "

C:\Users\F\Anaconda3\lib\site-packages\dask\array\core.py in normalize_chunks(chunks, shape)
   1815             raise ValueError(
   1816                 "Chunks and shape must be of the same length/dimension. "
-> 1817                 "Got chunks=%s, shape=%s" % (chunks, shape))
   1818 
   1819     if shape is not None:

ValueError: Chunks and shape must be of the same length/dimension. Got chunks=(5, 5), shape=(5000,)

问题是什么？

Answer 1

问题出在这里:

np.arange(n/2).reshape(n)

您创建了一个大小为 n/2 的数组，然后尝试将其 reshape 为大小为 n。您无法使用reshape更改大小。

这可能是复制/粘贴错误？它不在你的原始代码中，看来你正在做

np.arange(n).reshape(n/2,2)

在其他地方，只要 n 是偶数，它就可以工作(请小心，如果 n 不是偶数，这也会失败。)