BUG: Fix error for boxplot when using a pre-grouped DataFrame with more than one grouping
#57985
Conversation
| ret.loc[key] = d | ||
| # GH 14701 'key' needs to be converted to text as 'key' is a tuple when | ||
| # there is more than one group | ||
| ret.loc[pprint_thing(key)] = d |
There was a problem hiding this comment.
Instead of ret = pd.Series(dtype=object) above, can you do:
data = {}
for (key, group), ax in zip(grouped, axes):
...
data[key] = d
ret = pd.Series(data)
In general one should not construct a Series row-by-row (this is very slow), but rather provide all the data upfront. I think this will also allow tuples for key.
| ) | ||
| df["X"] = Series(np.repeat(["A", "B"], int(rows / 2))) | ||
| df["Y"] = Series(np.tile(["C", "D"], int(rows / 2))) | ||
| pregrouped = df.groupby(group) |
There was a problem hiding this comment.
nit: Just to adhere to existing conventions in tests, could you either call this grouped or gb. These are the names used in other tests here.
doc/source/whatsnew/v3.0.0.rst
Outdated
| @@ -375,7 +375,7 @@ Period | |||
|
|
|||
| Plotting | |||
| ^^^^^^^^ | |||
| - | |||
| - Bug in :meth:`DataFrameGroupBy.boxplot` passes a ``tuple`` instead of a ``string`` when input ``DataFrame`` is pre-grouped using more than one ``column`` (:issue:`14701`) | |||
There was a problem hiding this comment.
The link to the API docs won't resolve unless you include a . before DataFrameGroupBy; so it should be :meth:`.DataFrameGroupBy.boxplot`
Thanks for the feedback. I will make the necessary adjustments and update the PR accordingly. |
| # GH 14701 refactored to allow the 'key' to be passed as a tuple, | ||
| # which occurs when there is more than one group | ||
| data[key] = d | ||
| ret = pd.Series(data) |
There was a problem hiding this comment.
We don't need to be creating ret on every loop, only after the for loop is done. This line can be dedented.
There was a problem hiding this comment.
No problem.
| # GH 14701 refactored to allow the 'key' to be passed as a tuple, | ||
| # which occurs when there is more than one group |
There was a problem hiding this comment.
I think this comment should be removed.
doc/source/whatsnew/v3.0.0.rst
Outdated
| @@ -409,7 +409,7 @@ Period | |||
|
|
|||
| Plotting | |||
| ^^^^^^^^ | |||
| - | |||
| - Bug in :meth:`.DataFrameGroupBy.boxplot` passes a ``tuple`` instead of a ``string`` when input ``DataFrame`` is pre-grouped using more than one ``column`` (:issue:`14701`) | |||
There was a problem hiding this comment.
This is no longer accurate. I'd recommend something more generic, such as
Bug in
:meth:`.DataFrameGroupBy.boxplot`failed when there were multiple groupings (:issue:14701)
There was a problem hiding this comment.
Correct. I will update it.
|
Thanks @thetestspecimen - very nice! |
No Problem @rhshadrach. Thanks for the detailed input and guidance. |
…h more than one grouping (pandas-dev#57985)
doc/source/whatsnew/v3.0.0.rstfile if fixing a bug or adding a new feature.The original bug shows an indexing error when using
boxploton a pre-groupedDataFrame. For example:df.groupby(['X', 'Y']).boxplot()This error occurs as multiple groupings are passed as a tuple into the index of a
Series, which is not allowed. The fix converts thetupleto astring, which is acceptable as an index to aSeries.There are other comments on the bug thread that state "Groupby boxplot is not working even when a single 'by' is used", which is not the case.
I have included in the tests both a single
byand multiplebycovering three different methods of plotting aboxplotfrom aDataFrame:df.boxplot(by='X')df.boxplot(by=['X','Y'])df.plot.box(by='X')df.plot.box(by=['X','Y'])df.groupby('X').boxplot()df.groupby(['X','Y']).boxplot()Of all of the above, only example 6 required the fix.