FIX: handle fully masked data #9285
Conversation
|
probably worth adding in the comment that the actual value doesn't matter. |
lib/matplotlib/image.py
Outdated
| try: | ||
| a_max = a_max.astype(scaled_dtype) | ||
| except AttributeError: | ||
| a_min = 1 |
There was a problem hiding this comment.
simpler?
if np.ma.count(A) == 0:
a_min, a_max = 0, 1 # all masked, so values don't matter
else:
a_min = np.ma.min(A).astype(scaled_dtype)
a_max = np.ma.max(A).astype(scaled_dtype)
lib/matplotlib/image.py
Outdated
| a_max = np.ma.max(A) | ||
|
|
||
| # we need these try/except blocks to handle | ||
| # fully-masked/invalid input |
f5c69f6 to
659c9f3
Compare
|
Actually now that I think of it it may be slightly more efficient to check whether |
|
It is always masked because an earlier step in the processing chain makes it so. |
|
Then it definitely seems better to avoid a redundant full iteration... |
659c9f3 to
aa66aee
Compare
|
ok, so back to the original version of this PR. |
|
I would make the test on masked explicit rather than relying on attributeerror (but again no big deal) |
lib/matplotlib/image.py
Outdated
| except AttributeError: | ||
| a_min = 0 | ||
| try: | ||
| a_max = a_max.astype(scaled_dtype) |
There was a problem hiding this comment.
Gee, can't leave it alone.
try:
a_min = a_min.astype(scaled_dtype)
except AttributeError:
a_min, a_max = 0, 1
else:
a_max = a_max.astype(scaled_dtype)There was a problem hiding this comment.
Or shorten the whole block. From the top:
try:
a_min = A.min().astype(scaled_dtype)
except AttributeError:
a_min, a_max = 0, 1 # all masked, so values don't matter
else:
a_max = A.max().astype(scaled_dtype)There was a problem hiding this comment.
And taking @anntzer's suggestion:
a_min = A.min().astype(scaled_dtype)
if a_min is np.ma.masked:
a_min, a_max = 0, 1 # all masked, so values don't matter
else:
a_max = A.max().astype(scaled_dtype)|
I took the liberty of directly making a commit with a short form that also addresses @anntzer's suggestion instead of pestering you with yet another code chunk in a comment. The image code is sufficiently long and complicated that it is worthwhile to keep the handling of this special case as concise and readable as possible. |
|
@efiring take care of the merge conflict too? |
|
Yup, just did it. Tests are running. |
|
ah the joys of pep8 |
|
The problem there is the lousy editor available for in-place editing in github. It neither shows nor removes trailing whitespace. I will try again, but it will be a shot in the dark. If it doesn't work I guess I will have to go through the whole pull/modify/commit/push dance. |
As min/max do not make sense for an array with no non-masked values numpy returns a singleton, `np.ma.masked`, which can not be cast to a number. In this case just treat numbers as in range (0, 1) (even though it will just be ignored due to the masking). closes matplotlib#9280
a4cd618 to
63d3fb0
Compare
|
rebased and force-pushed to remove the merge commit. |
|
thanks |
As min/max do not make sense for an array with no non-masked values
numpy returns a singleton,
np.ma.masked, which can not be cast toa number. In this case just treat numbers as in range (0, 1) (even
though it will just be ignored due to the masking).
closes #9280
PR Summary
PR Checklist