It's nice to see that after some sleepless nights I managed to wrap up my first Kaggle competition on a positive note: jumped up 389 positions on the private leaderboard. Even though the final rank is 2067 (top 42%), I'm pretty stoked about my progress so far.
\nHere are a couple of things I've learned during this competition:
\nWhat I plan to learn from the top solutions once/if they become public:
\nThanks @cdeotte @raddar @ragnar123 @illidan7 @roberthatch for sharing your work and your findings!
\nOn to the next one!
", "messages": [ { "id": "1913676", "postDate": "08/25/2022 12:47:59", "content": "It's nice to see that after some sleepless nights I managed to wrap up my first Kaggle competition on a positive note: jumped up 389 positions on the private leaderboard. Even though the final rank is 2067 (top 42%), I'm pretty stoked about my progress so far.
\nHere are a couple of things I've learned during this competition:
\nWhat I plan to learn from the top solutions once/if they become public:
\nThanks @cdeotte @raddar @ragnar123 @illidan7 @roberthatch for sharing your work and your findings!
\nOn to the next one!
", "rawMarkdown": "It's nice to see that after some sleepless nights I managed to wrap up my first Kaggle competition on a positive note: jumped up 389 positions on the private leaderboard. Even though the final rank is 2067 (top 42%), I'm pretty stoked about my progress so far. \n\nHere are a couple of things I've learned during this competition:\n- How to work with memory limitations\n- How to work with RAPIDS\n- Feature engineering ideas (my favorite trick was to combine max and min into a difference of max - min feature; it achieved the same score as having max and min as separate sets of features freeing up almost 300 data columns)\n- I'm not so good at feature selection: I tried permutation importance, SHAP, Boruta, adversarial validation, tree feature importance and I couldn't get to a point where I felt comfortable with any of the results\n- DART boosting - a slow but quite accurate boosting algorithm\n- Ensembling works well: two weak models outscored my best single model\n- Model validation - I managed to handpick the best model and the private score went up which is a positive note for me\n\nWhat I plan to learn from the top solutions once/if they become public:\n - Knowledge distillation: seemed to have played an important role in this competition\n - Become better at feature engineering:\n \t- check impact of dimensionality reduction techniques (umap/PCA etc.)\n \t- find features that describe the 13-period time series more accurately (wasn't able to find the sweet spot during the competition)\n \t- sky is the limit here, or maybe the RAM size :)\n - Feature selection with [RAPIDS FIL](https://docs.rapids.ai/api/cuml/stable/api.html#forest-inferencing)\n\nThanks @cdeotte @raddar @ragnar123 @illidan7 @roberthatch for sharing your work and your findings!\n\nOn to the next one!", "votes": null }, { "id": "1913723", "postDate": "08/25/2022 13:18:55", "content": "This was a great challenge and an invaluable learning experience for me too! Many thanks for the public datasets and discussion posts that helped me a lot, special mention to @cdeotte @raddar @ragnar123 @illidan7 @roberthatch @ambrosm @tilli7 @thedevastator for their work!
", "rawMarkdown": "This was a great challenge and an invaluable learning experience for me too! Many thanks for the public datasets and discussion posts that helped me a lot, special mention to @cdeotte @raddar @ragnar123 @illidan7 @roberthatch @ambrosm @tilli7 @thedevastator for their work!", "votes": null }, { "id": "1914076", "postDate": "08/25/2022 17:51:42", "content": "My first time too! I had many of the same learnings :)
\nReally glad to hear my notebooks were of some benefit to you :)
", "rawMarkdown": "My first time too! I had many of the same learnings :)\n\nReally glad to hear my notebooks were of some benefit to you :)", "votes": null }, { "id": "1922173", "postDate": "09/01/2022 10:13:40", "content": "Hi @canonicalized, thanks for sharing! May I invite you to participate in this survey regarding your experience on Kaggle (10 min)? This is not a scam. We are a group of researchers at the City University of Hong Kong. The survey link is: https://cityuhk.questionpro.com/survey-of-kaggle-contestants
", "rawMarkdown": "Hi @canonicalized, thanks for sharing! May I invite you to participate in this survey regarding your experience on Kaggle (10 min)? This is not a scam. We are a group of researchers at the City University of Hong Kong. The survey link is: https://cityuhk.questionpro.com/survey-of-kaggle-contestants", "votes": null }, { "id": "2311555", "postDate": "06/21/2023 08:32:56", "content": "Hi there! My name is Arif Patel and I am a developer from Dubai. I'm Also A Newbie Here
", "rawMarkdown": "***Hi there! My name is Arif Patel and I am a developer from Dubai. I'm Also A Newbie Here***", "votes": null } ], "comments": [ { "id": 1913723, "author_name": "ravi20076", "author_url": "", "post_date": "08/25/2022 13:18:55", "content": "This was a great challenge and an invaluable learning experience for me too! Many thanks for the public datasets and discussion posts that helped me a lot, special mention to @cdeotte @raddar @ragnar123 @illidan7 @roberthatch @ambrosm @tilli7 @thedevastator for their work!
", "votes": null, "replies": [] }, { "id": 1914076, "author_name": "roberthatch", "author_url": "", "post_date": "08/25/2022 17:51:42", "content": "My first time too! I had many of the same learnings :)
\nReally glad to hear my notebooks were of some benefit to you :)
", "votes": null, "replies": [] }, { "id": 1922173, "author_name": "lystriving", "author_url": "", "post_date": "09/01/2022 10:13:40", "content": "Hi @canonicalized, thanks for sharing! May I invite you to participate in this survey regarding your experience on Kaggle (10 min)? This is not a scam. We are a group of researchers at the City University of Hong Kong. The survey link is: https://cityuhk.questionpro.com/survey-of-kaggle-contestants
", "votes": null, "replies": [] }, { "id": 2311555, "author_name": "arifpatelpduk", "author_url": "", "post_date": "06/21/2023 08:32:56", "content": "Hi there! My name is Arif Patel and I am a developer from Dubai. I'm Also A Newbie Here
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1913676": "It's nice to see that after some sleepless nights I managed to wrap up my first Kaggle competition on a positive note: jumped up 389 positions on the private leaderboard. Even though the final rank is 2067 (top 42%), I'm pretty stoked about my progress so far. \n\nHere are a couple of things I've learned during this competition:\n- How to work with memory limitations\n- How to work with RAPIDS\n- Feature engineering ideas (my favorite trick was to combine max and min into a difference of max - min feature; it achieved the same score as having max and min as separate sets of features freeing up almost 300 data columns)\n- I'm not so good at feature selection: I tried permutation importance, SHAP, Boruta, adversarial validation, tree feature importance and I couldn't get to a point where I felt comfortable with any of the results\n- DART boosting - a slow but quite accurate boosting algorithm\n- Ensembling works well: two weak models outscored my best single model\n- Model validation - I managed to handpick the best model and the private score went up which is a positive note for me\n\nWhat I plan to learn from the top solutions once/if they become public:\n - Knowledge distillation: seemed to have played an important role in this competition\n - Become better at feature engineering:\n \t- check impact of dimensionality reduction techniques (umap/PCA etc.)\n \t- find features that describe the 13-period time series more accurately (wasn't able to find the sweet spot during the competition)\n \t- sky is the limit here, or maybe the RAM size :)\n - Feature selection with [RAPIDS FIL](https://docs.rapids.ai/api/cuml/stable/api.html#forest-inferencing)\n\nThanks @cdeotte @raddar @ragnar123 @illidan7 @roberthatch for sharing your work and your findings!\n\nOn to the next one!", "1913723": "This was a great challenge and an invaluable learning experience for me too! Many thanks for the public datasets and discussion posts that helped me a lot, special mention to @cdeotte @raddar @ragnar123 @illidan7 @roberthatch @ambrosm @tilli7 @thedevastator for their work!", "1914076": "My first time too! I had many of the same learnings :)\n\nReally glad to hear my notebooks were of some benefit to you :)", "1922173": "Hi @canonicalized, thanks for sharing! May I invite you to participate in this survey regarding your experience on Kaggle (10 min)? This is not a scam. We are a group of researchers at the City University of Hong Kong. The survey link is: https://cityuhk.questionpro.com/survey-of-kaggle-contestants", "2311555": "***Hi there! My name is Arif Patel and I am a developer from Dubai. I'm Also A Newbie Here***" }, "source": "meta" }