gitignore

.gitignore

@@ -0,0 +1 @@
+.DS_Store

README.md

@@ -1,37 +1,3 @@
----
-license: apache-2.0
-tags:
-- tabular-learning
-size_categories:
-- 100B<n<1T
----
-TabRepo contains the predictions and metrics of 1530 models evaluated on 211 classification and regression datasets. 
-This allows to compare against state-of-the-art AutoML systems or random configurations by querying 
-precomputed results. We also store and expose model predictions so any ensembling strategy can also be benchmarked 
-cheaply by just querying precomputed results.
-
-We give scripts from our paper, [TabRepo: A Large Scale Repository of Tabular Model Evaluations and its AutoML Applications](https://arxiv.org/abs/2311.02971), so that one can reproduce all experiments that compare different models and portfolio
-strategies against state-of-the-art AutoML systems.
-
-The key features of the repo are:
-* 1530 models densely evaluated on 211 datasets with 3 distinct folds using 8-fold bagged CV
-  * `1530*211*3*8` = 7.75 million models
-* 330 GB of model predictions
-* code to compare methods against state-of-the-art AutoML systems and random model configurations
-* fast evaluations of any ensemble of models from table lookups with a few engineering tricks:
-  * fast metric evaluations with specific optimized cpp code (for instance to compute roc auc)
-  * efficient format that loads model evaluation on the fly with low memory footprint
-
-Feel free to check the github repository: https://github.com/autogluon/tabrepo to see documentation on how to use the dataset.
-
-If you find this work useful for you research, please cite the following:
-```
-@inproceedings{
-  tabrepo,
-  title={TabRepo: A Large Scale Repository of Tabular Model Evaluations and its Auto{ML} Applications},
-  author={David Salinas and Nick Erickson},
-  booktitle={AutoML Conference 2024 (ABCD Track)},
-  year={2024},
-  url={https://openreview.net/forum?id=03V2bjfsFC}
-}
-```
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d4188a95310582ef1968a209ca66e3d25594ec0d0241b47c921592fae4112e9
+size 1839