![\"model](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F7917824%2Fe24ffac54f3afb707f6b5274dadd8801%2Fcomparison.png?generation=1745956899373239&alt=media\")
My final submission is an ensemble of seven models:
\nFeature engineering for the tree models was inspired by the Wikipedia article on card counting.
\n
I thank @andreipaulavets for organizing the competition, @ravi20076 for drawing my attention to it, and @masayakawamata for his discussion contributions.
\n", "comments": [ { "id": 3191013, "author_name": "Alif Sathar", "author_url": "", "post_date": "2025-05-01T09:03:30.267000", "content": "This is an excellent ensemble strategy — really impressive how you balanced traditional regressors with tree-based models like CatBoost and LightGBM. The feature engineering inspired by card counting was a clever touch, especially considering the domain. The CV score improvements with the ensemble are clear, and the model comparison plots are super informative! Curious: did you consider stacking or blending the outputs more formally, or was it a simple average ensemble?
", "votes": 1, "replies": [ { "id": 3191021, "author_name": "AmbrosM", "author_url": "", "post_date": "2025-05-01T09:28:47.500000", "content": "Hi @alifsathar, I stacked the models with BayesianRidge as you can verify in the source code.
Awesome ensemble! Love the mix of models and the clever card-counting features. Super insightful work.
", "votes": 1, "replies": [] }, { "id": 3189855, "author_name": "Andrei Paulavets", "author_url": "", "post_date": "2025-04-29T21:28:17.193000", "content": "Great comparison, well done!
", "votes": 1, "replies": [] }, { "id": 3189815, "author_name": "Ravi Ramakrishnan", "author_url": "", "post_date": "2025-04-29T20:10:03.253000", "content": "Congrats for the 4th place finish! I am glad to play a small part of this success @ambrosm!
", "votes": 2, "replies": [] } ], "comment_count": 5, "messages": [ { "id": 3189813, "postDate": "2025-04-29T20:02:49.327Z", "content": "My final submission is an ensemble of seven models:
\nFeature engineering for the tree models was inspired by the Wikipedia article on card counting.
\n
I thank @andreipaulavets for organizing the competition, @ravi20076 for drawing my attention to it, and @masayakawamata for his discussion contributions.
\n", "rawMarkdown": "My final submission is an ensemble of seven models:\n- Linear regression with polynomial features of degree 4\n- Ridge regression with a Nystroem kernel approximation\n- KernelRidge with a polynomial kernel of degree 5\n- KernelRidge with a kernel composed of a polynomial kernel and an RBF kernel\n- Ridge regression with feature engineering based on permutation probabilities\n- CatBoost with feature engineering based on card counting and bust probabilities\n- LightGBM with feature engineering based on card counting and bust probabilities\n\nFeature engineering for the tree models was inspired by the Wikipedia article on [card counting](https://en.wikipedia.org/wiki/Card_counting).\n\n\n\nI thank @andreipaulavets for organizing the competition, @ravi20076 for drawing my attention to it, and @masayakawamata for his discussion contributions.\n\n[Source code is here](https://www.kaggle.com/code/ambrosm/blackjack-great-ensemble).", "votes": 9 }, { "id": 3191013, "postDate": "2025-05-01T09:03:30.267Z", "content": "This is an excellent ensemble strategy — really impressive how you balanced traditional regressors with tree-based models like CatBoost and LightGBM. The feature engineering inspired by card counting was a clever touch, especially considering the domain. The CV score improvements with the ensemble are clear, and the model comparison plots are super informative! Curious: did you consider stacking or blending the outputs more formally, or was it a simple average ensemble?
", "rawMarkdown": "This is an excellent ensemble strategy — really impressive how you balanced traditional regressors with tree-based models like CatBoost and LightGBM. The feature engineering inspired by card counting was a clever touch, especially considering the domain. The CV score improvements with the ensemble are clear, and the model comparison plots are super informative! Curious: did you consider stacking or blending the outputs more formally, or was it a simple average ensemble?", "votes": 1, "replies": [ { "id": 3191021, "postDate": "2025-05-01T09:28:47.500Z", "content": "Hi @alifsathar, I stacked the models with BayesianRidge as you can verify in the source code.
Awesome ensemble! Love the mix of models and the clever card-counting features. Super insightful work.
", "rawMarkdown": "Awesome ensemble! Love the mix of models and the clever card-counting features. Super insightful work.", "votes": 1 }, { "id": 3189855, "postDate": "2025-04-29T21:28:17.193Z", "content": "Great comparison, well done!
", "rawMarkdown": "Great comparison, well done!", "votes": 1 }, { "id": 3189815, "postDate": "2025-04-29T20:10:03.253Z", "content": "Congrats for the 4th place finish! I am glad to play a small part of this success @ambrosm!
", "rawMarkdown": "Congrats for the 4th place finish! I am glad to play a small part of this success @ambrosm!", "votes": 2 } ], "complete": true, "source": "live", "fetched_at": "2026-10-20T02:32:56Z", "competition_id": 94020, "competition_slug": "black-jack-smart-effect-of-removal-ml", "team_id": 13460618, "solution_url": "/competitions/black-jack-smart-effect-of-removal-ml/writeups/ambrosm-4-great-ensemble" }