Series[Timestamp].diff() should return Series[Timedelta]
#821
Comments
|
Thanks for the report. I think we have to add |
|
looking at series.diff documentation it looks like diff of Series[int] returns Series[float] and I made a test to confirm this. Current stubs returns Series[int]. I am not sure this is a correct behavior of pandas? |
Yes, it seems that's an error in the stub for |
|
We have to manage objects also from
Any other object people can think of before starting to code ? |
You'd have to look at the definition of You might want to look at how we handle Thanks for looking into this. It's non-trivial. |
|
For info, if it is ok, I'll need to create new classes in |
You shouldn't have to do this. Just follow the pattern used for |
|
Almost finished fixing this issue in https://github.com/mutricyl/pandas-stubs/tree/821-series_diff-timestamp2 however I have got an issue with > check(
assert_type(
pd.Series(
[datetime.datetime.now().date(), datetime.datetime.now().date()]
).diff(),
"TimedeltaSeries",
),
pd.Series,
pd.Timedelta,
)This test fails in pytest with: >>> pd.Series( [datetime.datetime.now().date(), datetime.datetime.now().date()] ).diff()
0 NaT
1 0 days
dtype: timedelta64[ns]Looking at if isinstance(actual, pd.Series):
value = actual.iloc[0]first item of Any advise on how to solve this particular case @Dr-Irv ? |
|
Very interesting use case where how we designed Do the following: Introduce a parameter to |
* Adding tests for #821 * finish to add tests from pd._types.S1 definition * start fixing stubs * date le lt ge gt fixing * new diff overloads * adding a few Never * fix: check function updated for series.diff * clean comments and update period test * cover windows specific RuntimeWarning * PeriodSeries.diff is OffsetSeries * remove comments * try tables 3.9.2 to fix macos github test --------- Co-authored-by: Laurent Mutricy <[email protected]>
|
#907 closed this, but I am still seeing issues with the code example above. I minimized the code into this example: # python bug.py & mypy bug.py & pyright bug.py
from __future__ import annotations
from typing import reveal_type
import pandas as pd
def fun(stamps: pd.Series[pd.Timestamp]) -> pd.Series[pd.Timedelta]:
reveal_type(stamps)
print(type(stamps[1]))
deltas = stamps.diff()
reveal_type(deltas)
print(type(deltas[1]))
return deltas
times = pd.Series([pd.Timestamp(0), pd.Timestamp(0)])
fun(times)
So runtime behavior is as expected. However, This is with version |
|
As a workaround, I could replace Before (working): times: pd.Series[pd.Timestamp] = df[TIME_COL]After (not working): times: TimestampSeries = df[TIME_COL]
|
That's the right workaround for your example
We had to introduce
The reason that works is that we can't track the types of columns within a
This is again related to how the type checker is handling the generic aspect of In general, whenever you refer to a column of a times = cast("TimestampSeries", df[TIME_COL]) |
So one cast needs to be necessary at least. Accepting that, it makes sense doing it as early as possible to catch errors in the processing. Thanks a lot! |
Describe the bug
Series[Timestamp].diff()returnsSeries[Timedelta]but stubs indicate it'sSeries[Timestamp]instead.To Reproduce
Output:
Please complete the following information:
mypy==1.7.1,pyright==1.1.337,pandas-stubs==2.1.1.230928Additional context
I tried both
Series[Timestamp]andTimestampSeriesdue to #673 (comment)Possibly related: #718
The text was updated successfully, but these errors were encountered: