Improved resampling checks

[BSchilperoort]

In this PR:

• A robust check for frequency mismatch between the data and calendar is implemented (closes #10).
• Removed pytest.fixture(autouse=True), it was used incorrectly, and lead to over-testing.
• Resampling is changed so if the user inputs a DataArray, they get a DataArray back.
  • This lead to a change in the definition of "bounds" of the resampling output. Instead of "interval" with dims (anchor_year, i_interval, bounds), there are now two coords (left_bound and right_bound) with dims (anchor_year, i_interval). This change was required to preserve the interval datetimes in the DataArray output.

Additionally, I made some small changes to the ruff configuration, the tests are checked now as well, but docstyles are ignored for that folder.

[BSchilperoort]

Hm. Turns out that there's a problem with the traintest notebook. I think we'll have to check the xarray/pandas input data for reserved coordinate/column names, e.g.:

anchor_year, i_interval, left_bound, right_bound, is_target (rename from "target").

An error should be raised if the data already contains one or more of these

[geek-yang]

The checks for frequency works for most of the cases, but not for monthly calendar, for instance

time_index = pd.date_range("20181001", "20211001", freq="2M")
df = pd.DataFrame(
    data={
        "data1": np.random.random(len(time_index)),
    },
    index=time_index,
)

cal = monthly_calendar(anchor="10-15", length="1M")
cal = cal.map_to_data(df)
lilio.resample(cal, df)

This will throw the following warning messages, which indeed have nothing to do with frequency, but only NaN values.

/home/yangliu/AI4S2S/lilio/lilio/utils.py:104: FutureWarning: Units 'M', 'Y' and 'y' do not represent unambiguous timedelta values and will be removed in a future version.
  return pd.Timedelta(data_freq)
/home/yangliu/miniconda3/envs/s2spy/lib/python3.10/site-packages/numpy/core/fromnumeric.py:3474: RuntimeWarning: Mean of empty slice.
  return _methods._mean(a, axis=axis, dtype=dtype,
/home/yangliu/miniconda3/envs/s2spy/lib/python3.10/site-packages/numpy/core/_methods.py:189: RuntimeWarning: invalid value encountered in double_scalars
  ret = ret.dtype.type(ret / rcount)
/home/yangliu/AI4S2S/lilio/lilio/utils.py:80: UserWarning: The input data could not fully cover the calendar's intervals. Intervals without available data will contain NaN values.
  warnings.warn(

As we have provided monthly_calendar in the shorthands, it would be nice to have check for it as well.

[BSchilperoort]

The checks for frequency works for most of the cases, but not for monthly calendar, for instance

Thanks for spotting this bug, Yang. It turns out that the Calendar's frequency is checked correctly, but the data's frequency is not! I'll modify the check to account for monthly-freq data inputs.

[sonarcloud[bot]]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
0 Code Smells

96.6% Coverage
0.0% Duplication