Question

设置scipy稀疏数组A[0,:] = np.ones()的第一行效果很好，
但是尝试使用A[:,0] = np.ones()设置第一列会引发ValueError。
这是scipy 1.5.2中的错误，还是找不到描述此内容的文档？

9月13日回答：这是一个已知的错误区域，请参见issues/10695 以及最新的scipy/sparse/_index.py。但是，我尚未对此进行A[:,0]的测试。

""" scipy sparse A[:,0] = ndarray ValueError """
# sparse A[0,:] = ndarray works, sparse A[:,0] = ndarray raises ValueError
# https://stackoverflow.com/search?q=[scipy] [sparse-matrix] ValueError  > 100

import numpy as np
from scipy import sparse
# import warnings
# warnings.simplefilter( "ignore", sparse.SparseEfficiencyWarning )

def versionstr():
    import numpy, scipy, sys
    return "versions: numpy %s  scipy %s  python %s " % (
        numpy.__version__, scipy.__version__ , sys.version.split()[0] )

print( versionstr() )  # 11 Sep 2020: numpy 1.19.2  scipy 1.5.2  python 3.7.6

#...........................................................................
n = 3
ones = np.ones( n )

for A in [
        np.eye(n),
        sparse.eye( n ).tolil(),
        sparse.eye( n ).tocsr(),
        sparse.eye( n ).tocsr(),
        ]:
    print( "\n-- A:", type(A).__name__, A.shape )
    print( "A[0,:] = ones" )
    A[0,:] = ones
    print( "A: \n", getattr( A, "A", A ))  # dense

        # first column = ones --
    if sparse.issparse( A ):
        A[:,0] = ones.reshape( n, 1 )   # ok
        A[:,0] = np.matrix( ones ).T    # ok
        A[range(n),0] = ones            # ok
    try:
        print( "A[:,0] = ones" )
        A[:,0] = ones                   # A dense ok, A sparse ValueError
    except ValueError as msg:
        print( "ValueError:", msg )
        # ValueError: cannot reshape array of size 9 into shape (3,1)

Answer 1

我可能会称它为错误，是的-这不是我期望的行为。在幕后，看起来这是由np.broadcast_arrays()驱动的，当填充数组密集时会调用它。此函数将1d数组视为2d（1，N）数组。我期望基于numpy切片的行为，如果大小正确，将使用1d数组而不进行广播。

列切片：

>>> np.broadcast_arrays(np.ones((3,1)), A[:,0].A)
[array([[1.],
       [1.],
       [1.]]), array([[1.],
       [0.],
       [0.]])]
>>> np.broadcast_arrays(np.ones((3,)), A[:,0].A)
[array([[1., 1., 1.],
       [1., 1., 1.],
       [1., 1., 1.]]), array([[1., 1., 1.],
       [0., 0., 0.],
       [0., 0., 0.]])]
>>> np.broadcast_arrays(np.ones((1, 3)), A[:,0].A)
[array([[1., 1., 1.],
       [1., 1., 1.],
       [1., 1., 1.]]), array([[1., 1., 1.],
       [0., 0., 0.],
       [0., 0., 0.]])]

行切片：

>>> np.broadcast_arrays(np.ones((3, )), A[0, :].A)
[array([[1., 1., 1.]]), array([[1., 0., 0.]])]
>>> np.broadcast_arrays(np.ones((3, 1)), A[0, :].A)
[array([[1., 1., 1.],
       [1., 1., 1.],
       [1., 1., 1.]]), array([[1., 0., 0.],
       [1., 0., 0.],
       [1., 0., 0.]])]
>>> np.broadcast_arrays(np.ones((1, 3)), A[0, :].A)
[array([[1., 1., 1.]]), array([[1., 0., 0.]])]

稀疏稀疏A [：，0] = ndarray ValueError

1 个答案: