Get the latest tech news
Show HN: TabPFN v2 – A SOTA foundation model for small tabular data
Tabular Prior-data Fitted Network, a tabular foundation model, provides accurate predictions on small data and outperforms all previous methods on datasets with up to 10,000 samples by a wide margin.
We compared TabPFN against state-of-the-art baselines, including tree-based methods (random forest 38, XGBoost (XGB) 7, CatBoost 9, LightGBM 8), linear models, support vector machines (SVMs) 39 and MLPs 34. Evaluation metrics include ROC AUC (area under the receiver operating characteristic curve; One-vs-Rest) and accuracy for classification, and R 2(coefficient of determination) and negative RMSE (root mean squared error) for regression. TabPFN represents a major change in tabular data modelling, leveraging ICL to autonomously discover a highly efficient algorithm that outperforms traditional human-designed approaches on datasets with up to 10,000 samples and 500 features.
Or read this on Hacker News
