BUG/API: make setitem-inplace preserve dtype when possible with PandasArray, IntegerArray, FloatingArray #39044
Conversation
…sArray, IntegerArray, FloatingArray
jreback
added
Dtype Conversions
| @@ -28,6 +32,31 @@ def dtype(request): | |||
| return PandasDtype(np.dtype(request.param)) | |||
|
|
|||
|
|
|||
| orig_setitem = pd.core.internals.Block.setitem | |||
There was a problem hiding this comment.
can you use monkeypatch instead?
There was a problem hiding this comment.
this does use monkeypatch. the monkeypatched method calls the original method
|
rebased + green |
docstring updated + green |
There was a problem hiding this comment.
Can you summarize which behaviours are changed?
Eg from the description I would suppose that
df = pd.DataFrame({'a': [1, 2, 3]})
df.loc[:, 'a'] = pd.array([3, 4, 5])
changed behaviour to preserve the original dtype, but I don't see that directly tested?
Does this need a whatsnew?
| @@ -193,7 +197,20 @@ class TestGetitem(base.BaseGetitemTests): | |||
|
|
|||
|
|
|||
| class TestSetitem(base.BaseSetitemTests): | |||
| pass | |||
| def test_setitem_series(self, data, full_indexer): | |||
There was a problem hiding this comment.
Can you indicate here why it is overriding the base class?
| if not data._mask.any(): | ||
| # GH#38896 like we do with ndarray, we set the values inplace | ||
| # but cast to the new numpy dtype | ||
| expected = pd.Series(data.to_numpy(data.dtype.numpy_dtype), name="data") |
There was a problem hiding this comment.
Why are we converting to the numpy dtype here? That's also not the original dtype?
|
am ok with this, @jorisvandenbossche if any addl comments. |
|
There are some questions in my last review above for which I was still waiting on an answer |
|
let's stick a pin in this until #39163 goes in; it will be easier to address outstanding questions/comments at that point |
|
mothballing until after #39163 |
jbrockmendel
added
the
Mothballed
jbrockmendel
removed
the
Mothballed
xref #38896 (doesn't close). In general
df[:, "A"] = footries to operate in-place before falling back to casting. This makes sure we do that whenfoois aPandasArray,IntegerArray, orFloatingArrayI guess could/should do the same for BooleanArray