Bug: Made it so that 0 was included in uint8 #14412
Conversation
|
Thx for the PR. pls rebase existing PR from next. |
| @@ -45,3 +45,4 @@ Bug Fixes | |||
|
|
|||
| - Bug in ``pd.concat`` where names of the ``keys`` were not propagated to the resulting ``MultiIndex`` (:issue:`14252`) | |||
| - Bug in ``MultiIndex.set_levels`` where illegal level values were still set after raising an error (:issue:`13754`) | |||
| - Bug in ``pd.to_numeric`` where it would not downcast a 0 to a uint8 (:issue:`14404`) | |||
There was a problem hiding this comment.
pls include #14401. Also, the issue is common for unsigned, not only for uint8.
Current coverage is 85.26% (diff: 100%)@@ master #14412 diff @@
==========================================
Files 140 140
Lines 50631 50634 +3
Methods 0 0
Messages 0 0
Branches 0 0
==========================================
+ Hits 43167 43173 +6
+ Misses 7464 7461 -3
Partials 0 0
|
| @@ -391,6 +391,14 @@ def test_downcast(self): | |||
| res = pd.to_numeric(data, downcast=downcast) | |||
| tm.assert_numpy_array_equal(res, expected) | |||
|
|
|||
| #check that 0 works as a unsigned downcast | |||
|
|
|||
| data = [0, 1, 2, 3] | |||
There was a problem hiding this comment.
check this for all of the downcast parameters (where it should be included)
There was a problem hiding this comment.
Did you want me to just edit everything between line 327 and line 368 in that test to include 0's in the arrays? Trying it before and after, it fails before and passes after. That being said if you don't want me to touch it for some reason it's relatively easy to make a new one. Also is there a reason that 340 and 341 is redundant?
|
Added a series of tests which applied a pandas series that included the minimum and maximum values of each dtype to a to_numeric downcast to see if it would downcast to the appropriate dtype. It did not add to uint64 but I'm not sure if there were plans to include uint64 so I'll open up a new issue. Also edited the whatsnew. |
| @@ -45,3 +45,4 @@ Bug Fixes | |||
|
|
|||
| - Bug in ``pd.concat`` where names of the ``keys`` were not propagated to the resulting ``MultiIndex`` (:issue:`14252`) | |||
| - Bug in ``MultiIndex.set_levels`` where illegal level values were still set after raising an error (:issue:`13754`) | |||
| - Bug in ``pd.to_numeric`` where it would not downcast a 0 properly. (:issue:`14401`) | |||
There was a problem hiding this comment.
say downcast='unsigned' as a kwargs is passed
| @@ -401,6 +401,32 @@ def test_downcast(self): | |||
| res = pd.to_numeric(data, downcast=downcast) | |||
| tm.assert_numpy_array_equal(res, expected) | |||
|
|
|||
| # check that the smallest and largest values in each integer type pass to each type. | |||
| int8 = [-128, 127] | |||
There was a problem hiding this comment.
don't do this explicty, use np.iinfo(dtype).min/max as these vary by platform
There was a problem hiding this comment.
actually on 2nd though, might be ok if appveyor passes
There was a problem hiding this comment.
can you do this in a loop instead (eg. a loop of tuples of the start,stop, dtype)
There was a problem hiding this comment.
I was actually thinking dictionary if that's okay. And I believe that the numpy types are standard across platforms but I'm not positive which is why I didn't bother with the np.info(dtype).min/max but for safeties sake I'll include it.
I also added a series of other tests that included making sure that a dtype shifted dtype once it reached the next value and once again edited the whatsnew.
| @@ -48,3 +48,6 @@ Bug Fixes | |||
| - Bug in ``MultiIndex.set_levels`` where illegal level values were still set after raising an error (:issue:`13754`) | |||
| - Bug in ``DataFrame.to_json`` where ``lines=True`` and a value contained a ``}`` character (:issue:`14391`) | |||
| - Bug in ``df.groupby`` causing an ``AttributeError`` when grouping a single index frame by a column and the index level (:issue`14327`) | |||
| - Bug in ``pd.to_numeric`` where it would not downcast a 0 to a uint8 (:issue:`14404`) | |||
| @@ -401,6 +401,62 @@ def test_downcast(self): | |||
| res = pd.to_numeric(data, downcast=downcast) | |||
| tm.assert_numpy_array_equal(res, expected) | |||
|
|
|||
| # check that the smallest and largest values in each integer type pass to each type. | |||
There was a problem hiding this comment.
this is still overly verbose. do something like.
checks = [('int8', 'integer', [np.iinfo(np.int8).min, np.iinfo(np.int8).max]),
.....
]
for dtype, downcast, min_max in checks:
....
git diff upstream/master | flake8 --diffDecided to restart. Sorry for the inconvenience. -> #14472