doc/Usage_tips.md

Usage tips

When to use LimiX

LimiX is designed for tabular datasets comprising fewer than 50,000 samples and fewer than 10,000 features. Larger datasets increase the model’s hardware requirements, while the performance gains compared to models such as XGBoost may be limited.

Intended use cases for LimiX

LimiX can be applied to both classification、 regression and missing value imputation tasks involving tabular data. Compared with methods such as XGBoost and CatBoost, it demonstrates faster training and inference speeds, while achieving superior model performance.

Limitations in the use of LimiX

When the dataset comprises a large number of samples, LimiX may require more hardware resources and exhibit lower performance compared to supervised models such as XGBoost.

Computational and time requirements

Leveraging its model architecture, LimiX attains higher processing speeds than conventional machine learning methods and does not necessitate the extensive GPU resources typically required by large language models.