说实话,我也不确定我是否得到了正确的结果。它似乎是不一致/破损的。其中一部分原因是由于形状不一致,但并非全部如此。有些数据似乎正在消失。
例如(请注意形状):
In [1]: import numpy as np
In [2]: x = np.zeros(1, dtype=np.dtype([('field', '<f8', (1, 2))]))
In [3]: y = x[0]['field'].copy()
In [4]: y[0] = 3
In [5]: y[1] = 4
---------------------------------------------------------------------------
IndexError Traceback (most recent call last)
<ipython-input-5-cba72439f97c> in <module>()
----> 1 y[1] = 4
IndexError: index 1 is out of bounds for axis 0 with size 1
In [6]: y[0][1] = 4
In [7]: x
Out[7]:
array([([[0.0, 0.0]],)],
dtype=[('field', '<f8', (1, 2))])
In [8]: y
Out[8]: array([[ 3., 4.]])
In [9]: x[0]['field'] = y
In [10]: x
Out[10]:
array([([[3.0, 0.0]],)],
dtype=[('field', '<f8', (1, 2))])
为了更容易理解,让我们将形状简化。
In [1]: import numpy as np
In [2]: x = np.zeros(1, dtype=np.dtype([('field', '<f8', 2)]))
In [3]: y = x[0]['field'].copy()
In [4]: y[0] = 3
In [5]: y[1] = 4
In [6]: x[0]['field'] = y
In [7]: x
Out[7]:
array([([3.0, 0.0],)],
dtype=[('field', '<f8', (2,))])
In [8]: y
Out[8]: array([ 3., 4.])
在这种情况下,数据的去向不清楚。但是,以一种能够存储数据的方式进行分配似乎很容易实现。
有几个选择:
In [9]: x['field'][0] = y
In [10]: x
Out[10]:
array([([3.0, 4.0],)],
dtype=[('field', '<f8', (2,))])
In [11]: x['field'] = y * 2
In [12]: x
Out[12]:
array([([6.0, 8.0],)],
dtype=[('field', '<f8', (2,))])
In [13]: x['field'][:] = y
In [14]: x
Out[14]:
array([([3.0, 4.0],)],
dtype=[('field', '<f8', (2,))])
In [15]: x[0]['field'][:] = y * 2
In [16]: x
Out[16]:
array([([6.0, 8.0],)],
dtype=[('field', '<f8', (2,))])
__setitem __()
的问题,因为x [0:]['field'] = ...
有效!即使使用非常高的索引x [0:999999]['field'] = ...
,它们也会被简单地忽略... - Saullo G. P. Castro