BUG: Fix of var method for complex arrays
#13177
Conversation
…array. Removed a typo from initial commit.
numpy/core/_methods.py
Outdated
| x = um.multiply(x, um.conjugate(x), out=x).real | ||
| else: | ||
| x = um.multiply(x, x, out=x) | ||
| x = um.multiply(x,um.conjugate(x),out=x).real |
There was a problem hiding this comment.
This isn't quite the solution suggested by @eric-wieser in #13110 (comment): rather the suggestion is to replace issubclass(arr.dtype.type, nt.complexfloating) with not issubclass(arr.dtype.type, (nt.floating, nt.integer)) (or something like it, I didn't check naming details), i.e., we do want to keep the optimization for numbers we know are not complex (because .conjugate makes a copy; if it were just a read-only view this would not be an issue, but that's a separate question).
There was a problem hiding this comment.
Thank you @mhvk I have made the necessary changes in my latest commit. not issubclass(arr.dtype.type, (nt.floating, nt.integer)) definitely helps. This ensures that optimisation for non complex numbers is kept intact.
numpy/core/_methods.py
Outdated
| @@ -115,7 +115,7 @@ def _var(a, axis=None, dtype=None, out=None, ddof=0, keepdims=False): | |||
| # Note that x may not be inexact and that we need it to be an array, | |||
| # not a scalar. | |||
| x = asanyarray(arr - arrmean) | |||
| if issubclass(arr.dtype.type, nt.complexfloating): | |||
| if not issubclass(arr.dtype.type, nt.integer): | |||
There was a problem hiding this comment.
This should include also nt.floating. And might as well swap the branches, i.e.,
if issubclass(arr.dtype.type, (nt.floating, nt.integer)):
x = um.multiply(x, x, out=x)
else:
x = um.multiply(x, um.conjugate(x), out=x).real
var and std methods for complex arrays
var and std methods for complex arraysvar and std methods of complex arrays
var and std methods of complex arraysvar and std methods of complex arrays
var and std methods of complex arraysvar method for complex arrays
|
A program using unittest framework should suffice, right? |
|
@qawbecrdteyf - there are test files under |
|
this seems like a good place |
|
The test cases added are right, aren't they? |
| @@ -216,6 +216,9 @@ def test_var(self): | |||
| assert_(np.isnan(np.var([]))) | |||
| assert_(w[0].category is RuntimeWarning) | |||
|
|
|||
| B = np.array([None, 0]) | |||
There was a problem hiding this comment.
The test is great. As a final suggestion, could you add a comment along the lines of
# Regression test for gh-13177, that object arrays are treated correctly.
B = np.array([1j, 0], dtype=object)
(Note: might as well simplify the definition...)
There was a problem hiding this comment.
I don't see that this was resolved
There was a problem hiding this comment.
strange, I could have sworn that I saw the comment there when I approved... Anyway, not too big a deal.
|
Failure is unrelated, so merging. |
Closes bug #13110
The earlier code had an incorrect optimisation for variance and that has been changed. Seems to be a very pervasive bug for complex number arrays.