{ "id": 357086, "title": "Tips and tricks to know for tree based model from Grandmaster", "url": "/competitions/tabular-playground-series-oct-2022/discussion/357086", "author_name": "Rashmi Margani", "post_date": "2022-10-03T07:03:20.696000", "votes": 11, "comment_count": 0, "views": 0, "content": "

1.XGBoost for Beginner by DanB

\n

2.Here is some useful note about using XGBoost on Rapid for using GPU faster Rapids xgboost tips by Yassine Alouini

\n

3.Good summary of XGBoost vs CatBoost vs LightGBM

\n

4.Some tricks for tuning parameters Tips and tricks to win kaggle data science competitions by Raddar.

\n

Note:

\n

Set learning rate (eta) parameter: 0.1-0.2 (based on dataset sizes and available resources)
\nTest maximum tree depth parameter, pick the best performing on CV
\nRegularization: Lamba, alpha
\nTune min leaf node size (aka min_child_weight)
\nTune randomness of each iteration (column/row sampling)
\nDecrease eta to value when you are comfortable with your hardware, like 0.025.
\nIf training time is reasonable, introduce bagging (num_parallel_trees)

", "messages": [ { "id": 1968670, "postDate": "2022-10-03T07:03:20.697Z", "content": "

1.XGBoost for Beginner by DanB

\n

2.Here is some useful note about using XGBoost on Rapid for using GPU faster Rapids xgboost tips by Yassine Alouini

\n

3.Good summary of XGBoost vs CatBoost vs LightGBM

\n

4.Some tricks for tuning parameters Tips and tricks to win kaggle data science competitions by Raddar.

\n

Note:

\n

Set learning rate (eta) parameter: 0.1-0.2 (based on dataset sizes and available resources)
\nTest maximum tree depth parameter, pick the best performing on CV
\nRegularization: Lamba, alpha
\nTune min leaf node size (aka min_child_weight)
\nTune randomness of each iteration (column/row sampling)
\nDecrease eta to value when you are comfortable with your hardware, like 0.025.
\nIf training time is reasonable, introduce bagging (num_parallel_trees)

", "rawMarkdown": "1.[XGBoost](https://www.kaggle.com/code/dansbecker/xgboost/notebook) for Beginner by DanB\n\n2.Here is some useful note about using [XGBoost on Rapid](https://www.kaggle.com/competitions/m5-forecasting-accuracy/discussion/151869#909931) for using GPU faster Rapids xgboost tips by Yassine Alouini\n\n3.[Good summary of XGBoost vs CatBoost vs LightGBM](https://www.kaggle.com/competitions/LANL-Earthquake-Prediction/discussion/89909)\n\n4.Some tricks for tuning parameters [Tips and tricks to win kaggle data science competitions](https://www.slideshare.net/DariusBaruauskas/tips-and-tricks-to-win-kaggle-data-science-competitions) by Raddar. \n\nNote:\n\nSet learning rate (eta) parameter: 0.1-0.2 (based on dataset sizes and available resources)\nTest maximum tree depth parameter, pick the best performing on CV\nRegularization: Lamba, alpha\nTune min leaf node size (aka min_child_weight)\nTune randomness of each iteration (column/row sampling)\nDecrease eta to value when you are comfortable with your hardware, like 0.025.\nIf training time is reasonable, introduce bagging (num_parallel_trees)\n\n\n", "votes": 10 } ], "comments": [], "raw_markdown_by_id": { "1968670": "1.[XGBoost](https://www.kaggle.com/code/dansbecker/xgboost/notebook) for Beginner by DanB\n\n2.Here is some useful note about using [XGBoost on Rapid](https://www.kaggle.com/competitions/m5-forecasting-accuracy/discussion/151869#909931) for using GPU faster Rapids xgboost tips by Yassine Alouini\n\n3.[Good summary of XGBoost vs CatBoost vs LightGBM](https://www.kaggle.com/competitions/LANL-Earthquake-Prediction/discussion/89909)\n\n4.Some tricks for tuning parameters [Tips and tricks to win kaggle data science competitions](https://www.slideshare.net/DariusBaruauskas/tips-and-tricks-to-win-kaggle-data-science-competitions) by Raddar. \n\nNote:\n\nSet learning rate (eta) parameter: 0.1-0.2 (based on dataset sizes and available resources)\nTest maximum tree depth parameter, pick the best performing on CV\nRegularization: Lamba, alpha\nTune min leaf node size (aka min_child_weight)\nTune randomness of each iteration (column/row sampling)\nDecrease eta to value when you are comfortable with your hardware, like 0.025.\nIf training time is reasonable, introduce bagging (num_parallel_trees)\n\n\n" } }