BUG: multifield-view of MaskedArray gets bad fill_value #12408
Conversation
3733bd0 to
27414ff
Compare
5dcdd1b to
204bd5d
Compare
|
So, removing that |
|
Very possible, but I'm not seeing it yet. Could you give an example? I see that numpy currently does (old behavior): >>> x = np.ma.ones(3, dtype=[('a','u2'), ('b','u4',(2,))])
>>> x['b'].fill_value
999999
>>> x[['b']].fill_value
(16959, [999999, 999999])This PR (new behavior): >>> x = np.ma.ones(3, dtype=[('a','u2'), ('b','u4',(2,))])
>>> x['b'].fill_value
999999
>>> x[['b']].fill_value
([999999, 999999],) |
|
Seems reasonable. |
numpy/ma/core.py
Outdated
| @@ -3245,7 +3244,8 @@ def _scalar_heuristic(arr, elem): | |||
| # Inherit attributes from self | |||
| dout._update_from(self) | |||
| # Check the fill_value | |||
| if isinstance(indx, basestring): | |||
| if isinstance(indx, basestring) or (isinstance(indx, list) and | |||
| builtins.all(isinstance(s, basestring) for s in indx)): | |||
There was a problem hiding this comment.
Possible candidate for a helper function, _is_field_index(idx)
There was a problem hiding this comment.
I originally did that, but later deleted it since it was only used in one place.
There was a problem hiding this comment.
Helper functions are still useful even if they get called only once - if you pick a good enough name, I can tell what it does without looking at the implementation - and if you don't, at least they provide a good place to put a docstring.
|
@ahaldane: here's an example when the behavior changes: My prediction is that this patch changes the second case to match the first, which is probably the right thing to do here. |
numpy/ma/core.py
Outdated
| elif fields: | ||
| fdtype = [(_[0], _[1]) for _ in ndtype.descr] | ||
| elif ndtype.names is not None: | ||
| fdtype = [(name, ndtype.fields[name][0]) for name in ndtype.names] |
There was a problem hiding this comment.
At this point, it's not clear to me why this line isn't just fdtype - ndtype
There was a problem hiding this comment.
True.. I just tried it, it passes all tests. So let's do that.
204bd5d to
1712430
Compare
| @@ -2029,6 +2029,16 @@ def test_fillvalue(self): | |||
| assert_equal(x.fill_value, 999.) | |||
| assert_equal(x._fill_value, np.array(999.)) | |||
|
|
|||
| # gh-10483 test multi-field index fill value | |||
There was a problem hiding this comment.
Might be worth having a separate test here
1712430 to
6386fdb
Compare
|
Fixed up following your suggestions, thanks! |
|
Thanks Allan. |
Fixes #10483
The problem was the fill_value was not being updated correctly for multifield-views of masked-arrays. Also,
dtype.descrwas used incorrectly. This eliminates the last use of.descrinnp.ma.