Scaling Backends : Polars, Hugging Face and Foundation Models
- Mentors
- Bene, Kiril Ralinovski
- Organization
- sktime
- Technologies
- python, Transformers, Polars
- Topics
- machine learning, deep learning, time-series, Forecasting
The idea behind this project is to integrate data backends like Polars and deep learning foundation models from libraries like Hugging-Face and building interfaces for them in sktime. Adding support to polars Series and DataFrame as mtypes which will represent sktime compatible scitypes will be a part of scaling backends for sktime datatypes. This includes writing/enhancing adapter for Polars, and implementing polars mtypes as an abstract data type, a scitype(Panel, Series). Adding Polars support to interfaced models that already support polars so users can directly pass Polars DataFrame for training and prediction in sktime.