4

Einstein summation (numpy.einsum) of boolean arrays in numpy doesn't produce expected results. Numpy.einsum function does logical operations on boolean arrays, which is questionable in the numeric contexts. 

# summation of a boolean numpy array

x = numpy.array([True, False, True])

print(numpy.sum(x))
# output: 2

print(numpy.einsum('i->', x))
# output: True

For a boolean array x = [True, False, True], I expect that the summation of x is 2, and the result should not depend on the particular choice of the function. However, numpy.sum gave 2, and numpy.einsum gave True.

I am not sure whether I misunderstood something or there is some problem with my code. Any help is appreciated.

ti7
  • 16,375
  • 6
  • 40
  • 68
yuyangtj
  • 43
  • 5
  • 1
    It's not obvious to me that 'sum' on a boolean array should produce a numeric result. Logical sum/and makes as much sense to me, maybe more. `x@np.array([True,True,True])` produces a boolean `True` as well. – hpaulj Jun 26 '19 at 16:20
  • it is my interpretation of a sentence appeared in python's documents, which says "In numeric contexts (for example when used as the argument to an arithmetic operator), they behave like the integers 0 and 1, respectively." (link https://docs.python.org/3.7/library/stdtypes.html). I am not 100% sure about the exact meaning of "numeric contexts". Not sure whether the following cases can be counted as the numeric contexts, `np.array([True, False])**np.array([True])`, `np.array([True, False])/np.array([True])`, and `np.array([True, False])*np.array([True])`. I think the question is too subjective – yuyangtj Jun 26 '19 at 20:14

1 Answers1

4

The difference here is that sum casts the boolean into integers before summing, while einsum skips this step except if you specify it explicitly.

Try:

print(numpy.einsum('i->', x, dtype=int))
Learning is a mess
  • 7,479
  • 7
  • 35
  • 71