切片类型影响numpy数组的可变性

Question

我用Python创建netCDF文件。当我尝试给变量的某些部分（或者叫切片）赋值时，能否成功赋值取决于切片的“类型”。

我不太明白这是为什么。任何帮助我理解这个问题的原因都会很感激。

例如：

import numpy as np
from netCDF4 import Dataset

nb_steps = 2
nb_lat = 3
nb_lon = 4

# open/create file
f = Dataset('/home/ccorbel/Desktop/test.nc', 'w', format='NETCDF3_CLASSIC')
f.createDimension('lat', nb_lat)
f.createDimension('lon', nb_lon)
f.createDimension('time', nb_steps)

# create/fill variables
variables = {}
variables['t'] = f.createVariable('temperature', 'float64', ('time', 'lat', 'lon'))
variables['t'][...] = np.zeros((nb_steps, nb_lat, nb_lon))

# "equivalent" to [0, :, ::-1]
slc  = [0, slice(None, None, None), slice(None, None, -1)]    

# "equivalent" to [0, :, :]
slc2 = [0, slice(None, None, None), slice(None, None, None)] 

# "equivalent" to [:, ::-1]
slc3 = [   slice(None, None, None), slice(None, None, -1)]

print type(variables['t'])
# type 'netCDF4.Variable'
print type(variables['t'][slc])
# type 'numpy.ndarray'
print type(variables['t'][slc][...])
# type 'numpy.ndarray'
print np.shape(variables['t'][slc])
# (3, 4)

# variables['t'][slc] = np.random.random((nb_lat, nb_lon))
# return IndexError: too many indices

variables['t'][slc][...] = np.random.random((nb_lat, nb_lon))
print '\n', variables['t'][...]

# [[[ 0.  0.  0.  0.]
#   [ 0.  0.  0.  0.]
#   [ 0.  0.  0.  0.]]
# 
#  [[ 0.  0.  0.  0.]
#   [ 0.  0.  0.  0.]
#   [ 0.  0.  0.  0.]]]

variables['t'][...] = np.zeros((nb_steps, nb_lat, nb_lon)) # reset
variables['t'][slc2] = np.random.random((nb_lat, nb_lon))[slc3]
print '\n', variables['t'][...]

# [[[ 0.17502009  0.98414122  0.89686025  0.11072791]
#   [ 0.51351626  0.09234043  0.54314083  0.937711  ]
#   [ 0.98732418  0.22714407  0.87387761  0.44653219]]

#  [[ 0.          0.          0.          0.        ]
#   [ 0.          0.          0.          0.        ]
#   [ 0.          0.          0.          0.        ]]]

variables['t'][...] = np.zeros((nb_steps, nb_lat, nb_lon)) # reset
#variables['t'][0, :, ::-1] = np.random.random((nb_lat, nb_lon)) 
# return IndexError: too many indices

variables['t'][0, :, ::-1][...] = np.random.random((nb_lat, nb_lon))
print '\n', variables['t'][...]

# [[[ 0.  0.  0.  0.]
#   [ 0.  0.  0.  0.]
#   [ 0.  0.  0.  0.]]

#  [[ 0.  0.  0.  0.]
#   [ 0.  0.  0.  0.]
#   [ 0.  0.  0.  0.]]]

variables['t'][...] = np.zeros((nb_steps, nb_lat, nb_lon)) # reset
variables['t'][0, :, :] = np.random.random((nb_lat, nb_lon))[:, ::-1]
print '\n', variables['t'][...]

# [[[ 0.61406835  0.11069783  0.28667398  0.45018246]
#   [ 0.3833354   0.98871281  0.55559104  0.60415683]
#   [ 0.75200954  0.75106639  0.11688565  0.14264615]]

#  [[ 0.          0.          0.          0.        ]
#   [ 0.          0.          0.          0.        ]
#   [ 0.          0.          0.          0.        ]]]

variables['t'][...] = np.zeros((nb_steps, nb_lat, nb_lon)) # reset
variables['t'][0, :, :] = np.random.random((nb_lat, nb_lon))[slc3]
print '\n', variables['t'][...]

# [[[ 0.09437484  0.45757906  0.81116891  0.23580254]
#   [ 0.37316425  0.06768454  0.20259876  0.42127472]
#   [ 0.78879307  0.62535419  0.08942293  0.68789143]]

#  [[ 0.          0.          0.          0.        ]
#   [ 0.          0.          0.          0.        ]
#   [ 0.          0.          0.          0.        ]]]

f.close()

Answer 1

你的示例代码在我这儿是可以运行的，但我觉得你可能遇到问题是因为你在赋值时使用了多个索引。比如说 A[0, :, ::-1][...] = something 这样的写法，其中 A 是一个数组，这种用法有点奇怪。虽然在我这儿看起来没问题，但我建议你尽量避免这样做。如果这样做还是不能解决你的问题，能不能给我们一个更简单的例子，最好是在 = 左边只用一个索引操作，或者解释一下你为什么想用两个索引操作。