TabArena
An open benchmark for tabular machine learning methods, covering datasets from multiple domains. It uses repeated random splits and statistical tests, separates single-model and ensemble/AutoML tracks, and reports cross-dataset ranking distributions.
Timeline (last 90 days)
Mitra-v2 leads the single-model track on the TabArena benchmark.
TabArena is an open evaluation benchmark for tabular machine learning methods, covering a large number of datasets from different domains, using multiple random splits and statistical tests to compare methods, and distinguishing between different tracks such as single-model and ensemble/AutoML.
Leading in the single-model track means achieving leading performance on a wide range of tasks with a single pretrained model, without relying on customizing a model for each dataset or ensembling multiple models.
Leading on a benchmark does not equal being superior on all real business tables; distribution differences between benchmark tasks and production data, as well as the choice of evaluation metrics, both affect the scope of extrapolating the conclusions.
All Facts (4)
Mitra-v2 leads the single-model track on the TabArena benchmark.
50%UnverifiedTabArena is an open evaluation benchmark for tabular machine learning methods, covering a large number of datasets from different domains, using multiple random splits and statistical tests to compare methods, and distinguishing between different tracks such as single-model and ensemble/AutoML.
50%UnverifiedLeading in the single-model track means achieving leading performance on a wide range of tasks with a single pretrained model, without relying on customizing a model for each dataset or ensembling multiple models.
50%UnverifiedLeading on a benchmark does not equal being superior on all real business tables; distribution differences between benchmark tasks and production data, as well as the choice of evaluation metrics, both affect the scope of extrapolating the conclusions.
50%