Closed
Description
On 1.11.1, random.normal does not preserve the shape of its inputs in one particular case: the size of the ndarray is one.
In[2]: import numpy as np
In[2]: from numpy.random import normal
Sending in Python lists all acts as expected:
In[4]: normal(0)
Out[4]: -0.5191599213417067
In[5]: normal([0])
Out[5]: array([-0.03654451])
In[6]: normal([[0]])
Out[6]: array([[-1.03033527]])
Numpy arrays of size 1 all return a float instead of an array of the corresponding shape:
In[10]: normal(np.asarray(0))
Out[10]: 0.6874599652327416
In[11]: normal(np.asarray([0]))
Out[11]: -0.044433450447185156
In[12]: normal(np.asarray([[0]]))
Out[12]: -0.4610284419029141
Arrays of other shapes work fine:
In[14]: normal(np.asarray([0, 0]))
Out[14]: array([-0.58998342, 0.13144979])
In[15]: normal(np.asarray([[0, 0]]))
Out[15]: array([[-0.14003459, 0.28130253]])
Metadata
Metadata
Assignees
Labels
No labels