discussionenhancementhelp wantedmodule: tabular
Repository metrics
- Stars
- (10,609 個のスター)
- PR merge metrics
- (平均マージ 4d 5h) (30d で 20 merged PRs)
説明
From tilii7's very nice kaggle post: https://www.kaggle.com/competitions/playground-series-s4e9/discussion/536980
- Investigate adding LAMA's NN model to AutoGluon.
- Investigate if hill climbing leads to performance improvement in ensembling. Can verify via TabRepo.
Both of these tasks are things that could be implemented by members in the community if they want to give it at try.
- For LAMA's NN: Refer to the custom model tutorial
- For hill climbing, refer to the experiment running code in TabRepo, you can replace the ensemble selection algorithm with a hill climbing algorithm and see if it improves the simulation score: https://github.com/autogluon/tabrepo/blob/main/scripts/baseline_comparison/evaluate_baselines.py