{ "id": 328801, "title": "Hopular with GBDT ....so XGBoost is all you need =)))", "url": "/competitions/amex-default-prediction/discussion/328801", "author_name": "", "post_date": "2022-06-03T04:40:09.654721900Z", "votes": 21, "comment_count": 1, "views": 0, "content": "

Recently, the name Hopular has become quite hot and here are a few lines from their Abstract Hopular: Modern Hopfield Networks for Tabular Data:

\n

In experiments on medium-sized tabular data with about 10,000 samples, Hopular outperforms XGBoost, CatBoost, LightGBM and a state-of-the art Deep Learning method designed for tabular data. Thus, Hopular is a strong alternative to these methods on tabular data.

\n

However, is it really as good as in paper 🤔🤔

\n

You can see more comment of @tunguz on his twitter Bojan Tunguz twitter and some of his quick comparison tests TabularBenchmarks Hopular with XGBoost

", "messages": [ { "id": "1809815", "postDate": "06/03/2022 04:40:09", "content": "

Recently, the name Hopular has become quite hot and here are a few lines from their Abstract Hopular: Modern Hopfield Networks for Tabular Data:

\n

In experiments on medium-sized tabular data with about 10,000 samples, Hopular outperforms XGBoost, CatBoost, LightGBM and a state-of-the art Deep Learning method designed for tabular data. Thus, Hopular is a strong alternative to these methods on tabular data.

\n

However, is it really as good as in paper 🤔🤔

\n

You can see more comment of @tunguz on his twitter Bojan Tunguz twitter and some of his quick comparison tests TabularBenchmarks Hopular with XGBoost

", "rawMarkdown": "Recently, the name Hopular has become quite hot and here are a few lines from their Abstract [Hopular: Modern Hopfield Networks for Tabular Data](https://arxiv.org/abs/2206.00664):\n\n**In experiments on medium-sized tabular data with about 10,000 samples, Hopular outperforms XGBoost, CatBoost, LightGBM and a state-of-the art Deep Learning method designed for tabular data. Thus, Hopular is a strong alternative to these methods on tabular data.**\n\nHowever, is it really as good as in paper 🤔🤔\n\nYou can see more comment of @tunguz on his twitter [Bojan Tunguz twitter](https://twitter.com/tunguz) and some of his quick comparison tests [TabularBenchmarks Hopular with XGBoost](https://github.com/tunguz/TabularBenchmarks/blob/main/scripts/college_HGB_KFold.ipynb)", "votes": null }, { "id": "1811011", "postDate": "06/04/2022 07:08:34", "content": "

Yeah you can't claim sota on 6 <500 rows data sets without a practical fit/predict interface for everyone to test.

", "rawMarkdown": "Yeah you can't claim sota on 6 <500 rows data sets without a practical fit/predict interface for everyone to test.", "votes": null } ], "comments": [ { "id": 1811011, "author_name": "lucasmorin", "author_url": "", "post_date": "06/04/2022 07:08:34", "content": "

Yeah you can't claim sota on 6 <500 rows data sets without a practical fit/predict interface for everyone to test.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1809815": "Recently, the name Hopular has become quite hot and here are a few lines from their Abstract [Hopular: Modern Hopfield Networks for Tabular Data](https://arxiv.org/abs/2206.00664):\n\n**In experiments on medium-sized tabular data with about 10,000 samples, Hopular outperforms XGBoost, CatBoost, LightGBM and a state-of-the art Deep Learning method designed for tabular data. Thus, Hopular is a strong alternative to these methods on tabular data.**\n\nHowever, is it really as good as in paper 🤔🤔\n\nYou can see more comment of @tunguz on his twitter [Bojan Tunguz twitter](https://twitter.com/tunguz) and some of his quick comparison tests [TabularBenchmarks Hopular with XGBoost](https://github.com/tunguz/TabularBenchmarks/blob/main/scripts/college_HGB_KFold.ipynb)", "1811011": "Yeah you can't claim sota on 6 <500 rows data sets without a practical fit/predict interface for everyone to test." }, "source": "meta" }