Numpy is absurd.

numpy_is_absurr.md

Numpy is absurd

HN Discussion

I always gripe about Python not having useful (i.e. performant and with adoption) built-in array type and Numpy doesn't distinguish "vector of vector" from "matrix", but this still surprised me.

in is actually intersect?

It seems that Numpy uses intersect logic to check a in b:

In [1]: import numpy as np

In [2]: np.__version__
Out[2]: '1.23.1'

In [3]: np.array([1]) in np.array([1,2,3])
Out[3]: True

In [4]: np.array([1,2,3]) in np.array([1])
Out[4]: True

also it doesn't care about dtype at all, so it's intersect but with value comparison:

In [5]: np.array([1,2,3,4,5]) in np.array([2.0])
Out[5]: True

In [6]: float(1) in np.array([1,2,3,4,5])
Out[6]: True

in is actually intersect(a.flat, b.flat)?

Things only get weired when we go to higher dimensional array:

In [7]: 2 in np.array([[1,2], [3, 4]])
Out[7]: True

In [8]: [2] in np.array([[1,2], [3, 4]])
Out[8]: True

In [9]: [[2]] in np.array([[1,2], [3, 4]])
Out[9]: True

In [10]: [[[2]]] in np.array([[1,2], [3, 4]])
Out[10]: True

In [11]: np.array([[[2]]]) in np.array([[1,2], [3, 4]])
Out[11]: True

Furthur more, because intersect is commutitive, in is also commutitive!

In [12]: np.array([[1,2], [3, 4]]) in np.array([2])
Out[12]: True

Wait, wtf

Finally, just when you throught you understand what Numpy is doing, it seems there's special treatment for ndarray of len() == 1

In [12]: np.array([1,2,3]) in np.array([2])
Out[12]: True

In [14]: np.array([1,2,3]) in np.array([2, 3])
<ipython-input-32-e5b43569579c>:1: DeprecationWarning: elementwise comparison failed; this will raise an error in the future.
  np.array([1,2,3]) in np.array([2, 3])
Out[14]: False

But then what is this?

In [15]: np.array([99,2,123]) in np.array([2, 30, 50])
Out[15]: False

In [16]: np.array([99,2,123]) in np.array([2, 123, 50])
Out[16]: False

In [17]: np.array([99,2,123]) in np.array([99, 123, 50])
Out[17]: True