{ "id": 368848, "title": "šŸ’” Training an XGBoost Ranker on the GPU with Merlin Models šŸ”„šŸ”„šŸ”„", "url": "/competitions/otto-recommender-system/discussion/368848", "author_name": "Radek Osmulski", "post_date": "2022-11-28T04:00:51.728000", "votes": 22, "comment_count": 2, "views": 0, "content": "

Hey!

\n

In order to ensemble predictions to improve our LB standing, we need to introduce diversity to our models.

\n

XGBoost is a great choice. We can train it on the GPU for extra speed. Also, whereas LGBM gives you just a single ranking objective to train with, XGBoost comes with 3!

\n\n

Training with different objectives can be a great way to improve the performance of your ensemble!

\n

Please find the notebook here:

\n

šŸ‘‰ šŸ† Training an XGBoost Ranker on the GPU šŸ”„šŸ”„šŸ”„

\n

Other resources you might find useful:

\n", "messages": [ { "id": 2046260, "postDate": "2022-11-28T04:00:51.730Z", "content": "

Hey!

\n

In order to ensemble predictions to improve our LB standing, we need to introduce diversity to our models.

\n

XGBoost is a great choice. We can train it on the GPU for extra speed. Also, whereas LGBM gives you just a single ranking objective to train with, XGBoost comes with 3!

\n\n

Training with different objectives can be a great way to improve the performance of your ensemble!

\n

Please find the notebook here:

\n

šŸ‘‰ šŸ† Training an XGBoost Ranker on the GPU šŸ”„šŸ”„šŸ”„

\n

Other resources you might find useful:

\n", "rawMarkdown": "Hey!\n\nIn order to ensemble predictions to improve our LB standing, we need to introduce diversity to our models.\n\n`XGBoost` is a great choice. We can train it on the GPU for extra speed. Also, whereas `LGBM` gives you just a single ranking objective to train with, XGBoost comes with 3!\n\n* `rank:pairwise`\n* `rank:ndcg`\n* `rank:map`\n\nTraining with different objectives can be a great way to improve the performance of your ensemble!\n\nPlease find the notebook here:\n\nšŸ‘‰ [šŸ† Training an XGBoost Ranker on the GPU šŸ”„šŸ”„šŸ”„](https://www.kaggle.com/radek1/training-an-xgboost-ranker-on-the-gpu)\n\n### Other resources you might find useful:\n\n* [šŸ’” [2 methods] How-to ensemble predictions šŸ…šŸ…šŸ…](https://www.kaggle.com/code/radek1/2-methods-how-to-ensemble-predictions)\n* [local validation tracks public LB perfecty -- here is the setup](https://www.kaggle.com/competitions/otto-recommender-system/discussion/364991)\n* [šŸ’” For my friends from Twitter and LinkedIn -- here is how to dive into this competition šŸ³](https://www.kaggle.com/competitions/otto-recommender-system/discussion/368560)\n* [Full dataset processed to CSV/parquet files with optimized memory footprint](https://www.kaggle.com/competitions/otto-recommender-system/discussion/363843)\n* [co-visitation matrix - simplified, imprvd logic šŸ”„](https://www.kaggle.com/code/radek1/co-visitation-matrix-simplified-imprvd-logic)\n* [šŸ’” Word2Vec How-to [training and submission]šŸš€šŸš€šŸš€](https://www.kaggle.com/code/radek1/word2vec-how-to-training-and-submission)\n", "votes": 20 }, { "id": 2052572, "postDate": "2022-12-02T10:00:40.733Z", "rawMarkdown": "", "votes": 1, "isDeleted": true, "replies": [ { "id": 2052575, "postDate": "2022-12-02T10:04:57.307Z", "content": "

Thank you, @cesaber! šŸ™Œ Your words to mean a lot to me, really appreciate you taking the time to share this feedback with me šŸ™‚šŸ™

", "rawMarkdown": "Thank you, @cesaber! šŸ™Œ Your words to mean a lot to me, really appreciate you taking the time to share this feedback with me šŸ™‚šŸ™", "votes": 1 } ] } ], "comments": [ { "id": 2052572, "author_name": "", "author_url": "", "post_date": "2022-12-02T10:00:40.733000", "content": "", "votes": 1, "replies": [ { "id": 2052575, "author_name": "Radek Osmulski", "author_url": "", "post_date": "2022-12-02T10:04:57.307000", "content": "

Thank you, @cesaber! šŸ™Œ Your words to mean a lot to me, really appreciate you taking the time to share this feedback with me šŸ™‚šŸ™

", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "2046260": "Hey!\n\nIn order to ensemble predictions to improve our LB standing, we need to introduce diversity to our models.\n\n`XGBoost` is a great choice. We can train it on the GPU for extra speed. Also, whereas `LGBM` gives you just a single ranking objective to train with, XGBoost comes with 3!\n\n* `rank:pairwise`\n* `rank:ndcg`\n* `rank:map`\n\nTraining with different objectives can be a great way to improve the performance of your ensemble!\n\nPlease find the notebook here:\n\nšŸ‘‰ [šŸ† Training an XGBoost Ranker on the GPU šŸ”„šŸ”„šŸ”„](https://www.kaggle.com/radek1/training-an-xgboost-ranker-on-the-gpu)\n\n### Other resources you might find useful:\n\n* [šŸ’” [2 methods] How-to ensemble predictions šŸ…šŸ…šŸ…](https://www.kaggle.com/code/radek1/2-methods-how-to-ensemble-predictions)\n* [local validation tracks public LB perfecty -- here is the setup](https://www.kaggle.com/competitions/otto-recommender-system/discussion/364991)\n* [šŸ’” For my friends from Twitter and LinkedIn -- here is how to dive into this competition šŸ³](https://www.kaggle.com/competitions/otto-recommender-system/discussion/368560)\n* [Full dataset processed to CSV/parquet files with optimized memory footprint](https://www.kaggle.com/competitions/otto-recommender-system/discussion/363843)\n* [co-visitation matrix - simplified, imprvd logic šŸ”„](https://www.kaggle.com/code/radek1/co-visitation-matrix-simplified-imprvd-logic)\n* [šŸ’” Word2Vec How-to [training and submission]šŸš€šŸš€šŸš€](https://www.kaggle.com/code/radek1/word2vec-how-to-training-and-submission)\n", "2052572": "" } }