Wrong documentation for Jaccard

in /spatial/distance.py, the documentation for the function jaccard specify that parameters u and v are both (N,) array_like, bool Input array. Howerver, in /spatial/tests/test_distance.py, there are function calls to jaccard where u and v are floating point valued arrays, am I missing something or should these tests not be there?

[rkern]

The _weight_checked() wrapper will implicitly change some of the values to be neither 0 nor 1.

In general, I think that many/most of these boolean-intended distances don’t deal with non-boolean values in a particularly consistent or useful manner. For instance, when we look at the example given for jaccard:

>>> distance.jaccard([1, 0, 0], [1, 1, 0])
0.5
>>> distance.jaccard([1, 0, 0], [1, 2, 0])
0.5

we might think, okay, we’re converting the values to booleans first by their truthiness (so 2 becomes True), but we’re not.

>>> distance.jaccard([1, 1, 0], [1, 2, 0])
0.5

The functions based on _nbool_correspond_all() also do weird things with non-bool values that I don’t think correspond with any well-known definition of these distances (but happy to be proved wrong, with accompanying tests for such defined behavior).

I think it would be a good idea to define (and test) the behavior of all of these functions that they will convert the input arrays to booleans by the truthiness of the inputs.

[aricooperdavis]

Ah I hadn’t seen that discussion - I agree that the functions should convert inputs to bool and be explicit about this in the documentation. I’ll remove this PR as the documentation can all be updated in #14357 so it’s updated at the same time as the behavior change?