Our solution is an ensemble of five different architectures, some manual empiric tuning and some seeds re-runs to help stabilizing the predictions and generalizing to the private test set.
\nEnsemble:
\nTotal: 35 models.
\nAll Hugging Face models were further pre-trained in the competition dataset, before the fine-tuning step.
\nSeed average helped improving ~0.002 on public LB score.
\nI want to thank some kagglers that were important to this score: @andretugan, @maunish, @rhtsingh, and @rajat95gupta. There were many others that helped one way or the other, by publicly sharing their works and thoughts.
\nSpecial thanks to my teammate @jbomitchell, who kept us updated with the most important public discussions and notebooks shared in this competition and also provided many valuable insights about the competition and train/test data.
\nFinally, thanks to everyone who joined, hosted or was a part of this competition!
\nCode is here.
", "comments": [ { "id": 1628426, "author_name": "Marília Prata", "author_url": "", "post_date": "2021-12-24T20:57:23.203000", "content": "Though it has already 5 months. Congrats for your medal and for being Competitions Expert.
\nTudo que depende apenas da gente tem muito mais valor. Parabéns HinePo.
Congrats on your silver medal, and becoming a Competitions Expert 🎉🎉🎉
", "votes": 1, "replies": [] }, { "id": 1420033, "author_name": "Rakesh Jarupula", "author_url": "", "post_date": "2021-08-03T02:55:45.640000", "content": "Congrats @hinepo 🎉🎉
", "votes": 2, "replies": [] }, { "id": 1418002, "author_name": "torch", "author_url": "", "post_date": "2021-08-03T00:57:53.130000", "content": "Many congratulations @hinepo and @jbomitchell . Thanks for sharing your solution and code.
", "votes": 2, "replies": [] }, { "id": 1417702, "author_name": "Miguel González González", "author_url": "", "post_date": "2021-08-03T00:35:32.460000", "content": "Congratulations for the silver! Thanks for sharing!
", "votes": 2, "replies": [] } ], "comment_count": 5, "messages": [ { "id": 1417457, "postDate": "2021-08-03T00:19:38.467Z", "content": "Our solution is an ensemble of five different architectures, some manual empiric tuning and some seeds re-runs to help stabilizing the predictions and generalizing to the private test set.
\nEnsemble:
\nTotal: 35 models.
\nAll Hugging Face models were further pre-trained in the competition dataset, before the fine-tuning step.
\nSeed average helped improving ~0.002 on public LB score.
\nI want to thank some kagglers that were important to this score: @andretugan, @maunish, @rhtsingh, and @rajat95gupta. There were many others that helped one way or the other, by publicly sharing their works and thoughts.
\nSpecial thanks to my teammate @jbomitchell, who kept us updated with the most important public discussions and notebooks shared in this competition and also provided many valuable insights about the competition and train/test data.
\nFinally, thanks to everyone who joined, hosted or was a part of this competition!
\nCode is here.
", "rawMarkdown": "Our solution is an ensemble of five different architectures, some manual empiric tuning and some seeds re-runs to help stabilizing the predictions and generalizing to the private test set.\n\nEnsemble:\n\n-\t2 roberta-base models, each one with a different head (5 fold CV each)\n-\tExtraTreesRegressor on the embeddings extracted from roberta-base (5 fold CV, 2 seeds)\n-\t1 roberta-large (5 fold)\n-\t1 Roberta-large Mean Pooling model (5 fold, 2 seeds)\n\nTotal: 35 models.\n\nAll Hugging Face models were further pre-trained in the competition dataset, before the fine-tuning step.\n\nSeed average helped improving ~0.002 on public LB score.\n\nI want to thank some kagglers that were important to this score: @andretugan, @maunish, @rhtsingh, and @rajat95gupta. There were many others that helped one way or the other, by publicly sharing their works and thoughts.\n\nSpecial thanks to my teammate @jbomitchell, who kept us updated with the most important public discussions and notebooks shared in this competition and also provided many valuable insights about the competition and train/test data.\n\nFinally, thanks to everyone who joined, hosted or was a part of this competition!\n\nCode is [here](https://www.kaggle.com/hinepo/clpr-5-models-v1-1).\n", "votes": 31 }, { "id": 1628426, "postDate": "2021-12-24T20:57:23.203Z", "content": "Though it has already 5 months. Congrats for your medal and for being Competitions Expert.
\nTudo que depende apenas da gente tem muito mais valor. Parabéns HinePo.
Congrats on your silver medal, and becoming a Competitions Expert 🎉🎉🎉
", "rawMarkdown": "Congrats on your silver medal, and becoming a Competitions Expert 🎉🎉🎉", "votes": 1 }, { "id": 1420033, "postDate": "2021-08-03T02:55:45.640Z", "content": "Congrats @hinepo 🎉🎉
", "rawMarkdown": "Congrats @hinepo 🎉🎉", "votes": 2 }, { "id": 1418002, "postDate": "2021-08-03T00:57:53.130Z", "content": "Many congratulations @hinepo and @jbomitchell . Thanks for sharing your solution and code.
", "rawMarkdown": "Many congratulations @hinepo and @jbomitchell . Thanks for sharing your solution and code. ", "votes": 2 }, { "id": 1417702, "postDate": "2021-08-03T00:35:32.460Z", "content": "Congratulations for the silver! Thanks for sharing!
", "rawMarkdown": "Congratulations for the silver! Thanks for sharing!", "votes": 2 } ], "complete": true, "source": "live", "fetched_at": "2026-10-19T23:43:41Z", "competition_id": 25914, "competition_slug": "commonlitreadabilityprize", "team_id": 6930522, "solution_url": "/competitions/commonlitreadabilityprize/writeups/hinepo-john-136th-place-solution-top-4-silver" }