Closed braaannigan closed 1 week ago
The primary issue is with polars, I've reported it here: https://github.com/pola-rs/polars/issues/16998
Could mlforecast pin numpy<2.0 for the moment?
We currently have polars only as a dev requirement (to run the tests) https://github.com/Nixtla/mlforecast/blob/2ec60c1435189da542fbcb5480e77b11c2bf584f/settings.ini#L18-L25
In my opinion the pin should come from polars, since that's where the incompatibility is at the moment.
polars added the pin in their numpy extra, so this should be fixed by using: pip install "mlforecast[polars]"
or pip install mlforecast "polars[numpy]"
. Feel free to reopen if you encounter this again.
What happened + What you expected to happen
Calling .fit or .cross_validaation gives a seg fault with numpy 2.0. The MRE below works with pandas but fails with Polars (try removing
pl.from_pandas
)Versions / Dependencies
polars==0.20.31 pandas==2.2.2 numba==0.60.0 numpy==2.0.0 (no issues when version is 1.26.4) mlforecast==0.13.0
Reproduction script
Issue Severity
High: It blocks me from completing my task.