where function with implicit booleans

[FrancescAlted]

From [email protected] on October 09, 2011 05:15:28

The following edge cases is not correctly handled by the numexpr 'where' function. Basically, implicit booleans (using 1 as True) are not handled as they are handled in numpy.

import numexpr as ne import numpy as np

np.where(1, 2, 3) array(2)

ne.evaluate('where(1, 2, 3)') Traceback (most recent call last): ... ValueError: too many dimensions

a = np.array([1]) b = 2 c = 3 np.where(a, b, c) array([2])

ne.evaluate('where(a, b, c)') Traceback (most recent call last): ... NotImplementedError: couldn't find matching opcode for 'where_llll'

Original issue: http://code.google.com/p/numexpr/issues/detail?id=61

[gdementen]

The first example in this issue was probably fixed by #287. Adding an explicit test for that would not hurt. The second example (which I guess is what was important to the original reporter) is probably not fixed.

[robbmcleod]

NE2.6 doesn't seem to cast the first argument in the ternary to bool if it's not that dtype. This is also true of NE3, so probably a check an operation insertion should be implemented. The default behavoir is 'safe' casting which definitely doesn't allow for int -> bool casts.

[github-actions[bot]]

Message to comment on stale issues. If none provided, will not mark issues stale