Error when computing multiple prediction intervals in Time Series Tutorial

[PolZT]

Hello MAPIE team,

Thanks for this wonderful library. I'm encountering an issue when trying to compute two intervals instead of one in the time series tutorial, but I might be wrong or missing something in my understanding.

Describe the bug

When attempting to compute two intervals instead of one by changing the alpha value from 0.05 to [0.1, 0.05] in the time series tutorial, a ValueError is raised. The error message indicates an issue with evaluating the truth value of an array in the conformity_scores.py file.

To Reproduce

Steps to reproduce the behavior:

1. Go to the time series tutorial provided in the MAPIE documentation: MAPIE Time Series Tutorial.
2. Replace the original alpha = 0.05 with alpha = [0.1, 0.05] on line 193 of the tutorial script.
3. Run the script.
4. See the error.

Expected behavior

The script should compute and return two prediction intervals without raising any errors.

Screenshots

N/A (the error message is provided below).

Desktop:

• OS: Windows 10
• IDE: PyCharm 2024.1 (Community Edition)
• MAPIE Version: 0.8.6

Additional context

I tried to find out what is causing it and here are some of my findings:

In conformity_scores.py, the get_quantile function (line 213) receives an inconsistent formatting of alpha_np, sometimes receiving a simple NDArray and sometimes receiving an array of arrays.

Error traceback:

y_pred_enbpi_npfit, y_pis_enbpi_npfit = mapie_enbpi.predict(
  File "C:\Users\pzaho\anaconda3\envs\dynamo\lib\site-packages\mapie\regression\time_series_regression.py", line 459, in predict
    return super().predict(
  File "C:\Users\pzaho\anaconda3\envs\dynamo\lib\site-packages\mapie\regression\regression.py", line 645, in predict
    self.conformity_score_function_.get_bounds(
  File "C:\Users\pzaho\anaconda3\envs\dynamo\lib\site-packages\mapie\conformity_scores\conformity_scores.py", line 454, in get_bounds
    quantile_low = self.get_quantile(
  File "C:\Users\pzaho\anaconda3\envs\dynamo\lib\site-packages\mapie\conformity_scores\conformity_scores.py", line 272, in get_quantile
    quantile = signed * np.column_stack([
  File "C:\Users\pzaho\anaconda3\envs\dynamo\lib\site-packages\mapie\conformity_scores\conformity_scores.py", line 276, in <listcomp>
    ) if not (unbounded and _alpha >= 1) else np.inf * np.ones(n_ref)
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

[Valentin-Laurent]

Hello @PolZT, thank you for reporting this, and your positive feedback.
We're working on it :)

[Valentin-Laurent]

Hello @PolZT.
While invesestigating your issue, I realized that the optimize_beta feature is not working as expected (see issue #588). This explains the error you get in the notebook.
We're currently discussing the priority of this fix VS other ongoing developments. Contributions are welcome!