BUG: Fix range dtype in Series/DataFrame constructor on Windows #17840
Conversation
pandas/core/series.py
Outdated
| elif isinstance(data, range): | ||
| # GH 16804 | ||
| start, stop, step = _get_range_parameters(data) | ||
| if is_extension_type(dtype): |
There was a problem hiding this comment.
What case does this handle? Doesn't seem to be a matching test
There was a problem hiding this comment.
This is for dtypes that are specific to pandas that numpy doesn't understand, e.g. pd.Series(range(5), dtype='category'). Will add a test for this when I fix the failing tests.
gfyoung
added
Bug
Compat
pandas/core/series.py
Outdated
| elif isinstance(data, range): | ||
| # GH 16804 | ||
| start, stop, step = _get_range_parameters(data) | ||
| if is_extension_type(dtype): |
There was a problem hiding this comment.
can you construct the range immediately after _get_range_parameters. I think this might just work then (you still prob need the _try_cast); I don't think its a big deal to always construct the range in int64, then cast if needed.
pandas/core/common.py
Outdated
| @@ -539,6 +539,30 @@ class Sentinel(object): | |||
| return Sentinel() | |||
|
|
|||
|
|
|||
| def _get_range_parameters(data): | |||
| """Gets the start, stop, and step parameters from a range object""" | |||
There was a problem hiding this comment.
maybe put this in pandas/compat/__init__ instead
pandas/tests/test_resample.py
Outdated
| @@ -3030,7 +3030,7 @@ def test_nearest(self): | |||
| # GH 17496 | |||
| # Resample nearest | |||
| index = pd.date_range('1/1/2000', periods=3, freq='T') | |||
| result = pd.Series(range(3), index=index).resample('20s').nearest() | |||
| result = pd.Series(np.arange(3), index=index).resample('20s').nearest() | |||
There was a problem hiding this comment.
so I would actually leave some tests with pure range in the constructor, e.g. is there a reason to change this one?
There was a problem hiding this comment.
The expected is pd.Series(np.array([0, ...]), index=...), and on Windows the np.array defaults to int32 dtype, so the test fails since result is now int64 dtype.
Either need range -> np.arange or np.array -> list (or specify the dtype in np.array) for the test to pass. Changed it to keep range here.
jschendel
force-pushed
the
fix-range-dtype
branch
from
77e986c to
58adb2e
Compare
Codecov Report
@@ Coverage Diff @@
## master #17840 +/- ##
=========================================
Coverage ? 91.22%
=========================================
Files ? 163
Lines ? 50112
Branches ? 0
=========================================
Hits ? 45713
Misses ? 4399
Partials ? 0
Continue to review full report at Codecov.
|
pandas/compat/__init__.py
Outdated
| @@ -146,6 +150,24 @@ def bytes_to_str(b, encoding='ascii'): | |||
| def signature(f): | |||
| return inspect.getargspec(f) | |||
|
|
|||
| def _get_range_parameters(data): | |||
There was a problem hiding this comment.
you can de-privatize these (e.g. remove the leading _)
jschendel
force-pushed
the
fix-range-dtype
branch
from
58adb2e to
dfa4573
Compare
|
ping @jreback : green |
|
thanks @jschendel ! |
rangein DataFrame constructor on Windows #16804git diff upstream/master -u -- "*.py" | flake8 --diffHad to modify a few existing tests that use
range, as this caused them to fail due to dtypes mismatch; mostly just involved switching tonp.arange.