Giters

functime-org / functime

Time-series machine learning at scale. Built with Polars for embarrassingly parallel feature extraction and forecasts on panel data.

Home Page:https://docs.functime.ai

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Reduced conversion precision prevents model fitting

jhug12 opened this issue · comments

commented

Describe the bug
The use of float32 precision in the X_to_numpy and y_to_numpy functions leads to numerical problems when fitting the elastic_net_cv forecaster. This reduced precision causes a ValueError in the sklearn api when computing the gram matrix

To Reproduce
Tried unsuccessfully to reproduce it using random data.

**Desktop **

  • OS: Windows 10
  • Python: 3.10.0
  • functime: 0.9.5

Additional context
This issue seems related to the general precision problem discussed in scikit-learn#21997. A potential enhancement could be to allow configuration of precision level through an API option, improving the flexibility and applicability of the model in diverse scenarios.