DOC: enable docstring on DataFrame.columns/index #20385
Conversation
pandas/_libs/properties.pyx
Outdated
| Py_ssize_t axis | ||
|
|
||
| def __init__(self, axis=0): | ||
| cdef readonly: | ||
| object __doc__ |
There was a problem hiding this comment.
Can this go under the same def readonly block as axis?
And do we want object or str for the type?
There was a problem hiding this comment.
whoops, the above cdef readonly is a typo, I meant to keep it as readonly.
I am not fully sure if it is actually needed for axis, or if we can also make it readonly (I would think it is never modified internally, bot not sure)
For the object, it is what is done for CachedProperties above as well. My cython is not good enough to really know the consequences. I think we usually use object for strings in our cython code.
There was a problem hiding this comment.
If the test pass with using readonly for axis, I will combine them in one cdef readonly
There was a problem hiding this comment.
object vs str is pretty harmless in this case. I'd advocate sticking with object for now b/c then we don't need to worry about potential py2/py3 bytes/unicode corner cases.
Codecov Report
@@ Coverage Diff @@
## master #20385 +/- ##
==========================================
+ Coverage 91.77% 91.8% +0.02%
==========================================
Files 152 152
Lines 49205 49215 +10
==========================================
+ Hits 45159 45181 +22
+ Misses 4046 4034 -12
Continue to review full report at Codecov.
|
pandas/core/generic.py
Outdated
| @@ -278,7 +278,7 @@ def _setup_axes(cls, axes, info_axis=None, stat_axis=None, aliases=None, | |||
| if build_axes: | |||
|
|
|||
| def set_axis(a, i): | |||
| setattr(cls, a, properties.AxisProperty(i)) | |||
| setattr(cls, a, properties.AxisProperty(i, a)) | |||
There was a problem hiding this comment.
This setattr has always bothered me. It kind of makes since in a general-case world with PanelND, but if we're only interested in Series/DataFrame/[barely-]Panel wouldn't it be clearer just to write:
class DataFrame(NDFrame):
index = properties.AxisProperty(0)
columns = properties.AxisProperty(1)
?
There was a problem hiding this comment.
Yep, fully agree that if we only have 1D/2D this makes much more sense. But maybe we should only clean that up when Panel is removed?
Follow-up on #18196, similar to #19991
Didn't add an actual full docstring yet, but this already gives:
instead of (on master, on 0.22 this raises an exception):