Fix/operand error with encoders #2034
Conversation
…ts a ambiguous timedelta value to extract the start time index
Codecov ReportAttention: ❗ Your organization needs to install the Codecov GitHub app to enable full functionality.
... and 6 files with indirect coverage changes 📢 Thoughts on this report? Let us know!. |
There was a problem hiding this comment.
Very nice, thanks a lot @madtoinou.
Just had a minor suggestion and that we should add a test for it
darts/utils/data/tabularization.py
Outdated
| start_time_idx = ( | ||
| len( | ||
| pd.date_range( | ||
| start=time_index_i[0], |
There was a problem hiding this comment.
It might be more efficient to generate the index from the end of the series instead of from the beginning and then just add the len(time_index_i) to it?
Also we could use our darts.utils.timeseries_generation.generate_index for that
There was a problem hiding this comment.
Yup, I don't know if this is much faster but at least, it looks similar to the other case
| @@ -1132,37 +1132,44 @@ def test_lagged_training_data_extend_past_and_future_covariates_range_idx(self): | |||
| assert np.allclose(expected_X, X[:, :, 0]) | |||
| assert np.allclose(expected_y, y[:, :, 0]) | |||
|
|
|||
| def test_lagged_training_data_extend_past_and_future_covariates_datetime_idx(self): | |||
| @pytest.mark.parametrize("freq", ["D", "MS", "Y"]) | |||
Fixes #1875, fixes #1991
Summary
When encoders are used to generate covariates, they have the minimum time requirements. In tabularization, an arithmetic operation on
Timedeltaandpandas.offsetmust be performed to realign the covariate and target time indexes. However, some frequencies ('M', 'Y' and 'y') conversion toTimedeltaare ambiguous (pandas doc), causing the unsupported operand error.To solve the problem for these specific cases, a temporary
DatetimeIndexis created and the information is extracted without relying on the conversion (slower than the arithmetic operation).