Congratulations to the winners, everyone else for contributing in one way or another to our lively and fun community here. I really enjoyed this competition and all the sharing by the community.
\n\nI am sharing the stage-2 public LB that is based on only 1% i.e. 30 images of the stage-2 test data. Out of nearly 1445 teams in stage-1, only 380 submitted in the 2nd stage.
\n\nI am happy to have chosen my final submission based on my local CV which turned out to be my best on the private LB. Even though the stage-2 public LB is not a good representation of the full stage-2 test data, I observed similar higher scores with models that had lower CV as I saw on the stage-1 LB.
\n\nWhat do you think? Did you choose wisely?
\n\nShare in the comments section below.
\n\nGood luck everyone and happy kaggling.
", "messages": [ { "id": "413466", "postDate": "11/01/2018 02:11:25", "content": "Congratulations to the winners, everyone else for contributing in one way or another to our lively and fun community here. I really enjoyed this competition and all the sharing by the community.
\n\nI am sharing the stage-2 public LB that is based on only 1% i.e. 30 images of the stage-2 test data. Out of nearly 1445 teams in stage-1, only 380 submitted in the 2nd stage.
\n\nI am happy to have chosen my final submission based on my local CV which turned out to be my best on the private LB. Even though the stage-2 public LB is not a good representation of the full stage-2 test data, I observed similar higher scores with models that had lower CV as I saw on the stage-1 LB.
\n\nWhat do you think? Did you choose wisely?
\n\nShare in the comments section below.
\n\nGood luck everyone and happy kaggling.
", "rawMarkdown": "Congratulations to the winners, everyone else for contributing in one way or another to our lively and fun community here. I really enjoyed this competition and all the sharing by the community.\n\nI am sharing the [stage-2 public LB][1] that is based on only 1% i.e. 30 images of the stage-2 test data. Out of nearly 1445 teams in stage-1, only 380 submitted in the 2nd stage.\n\nI am happy to have chosen my final submission based on my local CV which turned out to be my best on the private LB. Even though the stage-2 public LB is not a good representation of the full stage-2 test data, I observed similar higher scores with models that had lower CV as I saw on the stage-1 LB.\n\nWhat do you think? Did you choose wisely?\n\nShare in the comments section below.\n\nGood luck everyone and happy kaggling.\n\n\n [1]: https://www.kaggle.com/sheriytm/get-yagana-s-stage1-2-pdf-kaggle-bug-workaround/output", "votes": null }, { "id": "413475", "postDate": "11/01/2018 02:29:53", "content": "Congrats to all the winners! Can't wait to see the solutions.
\n\nI must admit to relying on the leaderboard for validation, since my final solution was trained on the whole dataset. I did end up picking my best performing private LB solution but I think luck had a part to play.
", "rawMarkdown": "Congrats to all the winners! Can't wait to see the solutions.\n\nI must admit to relying on the leaderboard for validation, since my final solution was trained on the whole dataset. I did end up picking my best performing private LB solution but I think luck had a part to play.", "votes": null }, { "id": "413566", "postDate": "11/01/2018 06:26:24", "content": "I was saved by the outage of Google Cloud last night. I could not generate the final submission with the new model trained on the whole dataset. I ended up using the earlier submission (from stage 1 model).\nHad GCP been up, I would have chosen that final submission due to higher LB score and I would be kicking myself today. \nLady luck is in play
", "rawMarkdown": "I was saved by the outage of Google Cloud last night. I could not generate the final submission with the new model trained on the whole dataset. I ended up using the earlier submission (from stage 1 model).\nHad GCP been up, I would have chosen that final submission due to higher LB score and I would be kicking myself today. \nLady luck is in play", "votes": null }, { "id": "413590", "postDate": "11/01/2018 06:59:10", "content": "Temptation!!! Glad Google cloud saved you :-)
\n\n@yukiya, that is why I posted this thread. I was really tempted but did not choose my highest submission which is the same Stage-1 final model but retrained to a different number of epochs on stage-2 data. The CV looked like it was over-fitting hence I did not choose it and I was right.
", "rawMarkdown": "Temptation!!! Glad Google cloud saved you :-)\n\n@yukiya, that is why I posted this thread. I was really tempted but did not choose my highest submission which is the same Stage-1 final model but retrained to a different number of epochs on stage-2 data. The CV looked like it was over-fitting hence I did not choose it and I was right.", "votes": null } ], "comments": [ { "id": 413475, "author_name": "lextoumbourou", "author_url": "", "post_date": "11/01/2018 02:29:53", "content": "Congrats to all the winners! Can't wait to see the solutions.
\n\nI must admit to relying on the leaderboard for validation, since my final solution was trained on the whole dataset. I did end up picking my best performing private LB solution but I think luck had a part to play.
", "votes": null, "replies": [] }, { "id": 413566, "author_name": "nyleve", "author_url": "", "post_date": "11/01/2018 06:26:24", "content": "I was saved by the outage of Google Cloud last night. I could not generate the final submission with the new model trained on the whole dataset. I ended up using the earlier submission (from stage 1 model).\nHad GCP been up, I would have chosen that final submission due to higher LB score and I would be kicking myself today. \nLady luck is in play
", "votes": null, "replies": [ { "id": 413590, "author_name": "sheriytm", "author_url": "", "post_date": "11/01/2018 06:59:10", "content": "Temptation!!! Glad Google cloud saved you :-)
\n\n@yukiya, that is why I posted this thread. I was really tempted but did not choose my highest submission which is the same Stage-1 final model but retrained to a different number of epochs on stage-2 data. The CV looked like it was over-fitting hence I did not choose it and I was right.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "413466": "Congratulations to the winners, everyone else for contributing in one way or another to our lively and fun community here. I really enjoyed this competition and all the sharing by the community.\n\nI am sharing the [stage-2 public LB][1] that is based on only 1% i.e. 30 images of the stage-2 test data. Out of nearly 1445 teams in stage-1, only 380 submitted in the 2nd stage.\n\nI am happy to have chosen my final submission based on my local CV which turned out to be my best on the private LB. Even though the stage-2 public LB is not a good representation of the full stage-2 test data, I observed similar higher scores with models that had lower CV as I saw on the stage-1 LB.\n\nWhat do you think? Did you choose wisely?\n\nShare in the comments section below.\n\nGood luck everyone and happy kaggling.\n\n\n [1]: https://www.kaggle.com/sheriytm/get-yagana-s-stage1-2-pdf-kaggle-bug-workaround/output", "413475": "Congrats to all the winners! Can't wait to see the solutions.\n\nI must admit to relying on the leaderboard for validation, since my final solution was trained on the whole dataset. I did end up picking my best performing private LB solution but I think luck had a part to play.", "413566": "I was saved by the outage of Google Cloud last night. I could not generate the final submission with the new model trained on the whole dataset. I ended up using the earlier submission (from stage 1 model).\nHad GCP been up, I would have chosen that final submission due to higher LB score and I would be kicking myself today. \nLady luck is in play", "413590": "Temptation!!! Glad Google cloud saved you :-)\n\n@yukiya, that is why I posted this thread. I was really tempted but did not choose my highest submission which is the same Stage-1 final model but retrained to a different number of epochs on stage-2 data. The CV looked like it was over-fitting hence I did not choose it and I was right." }, "source": "meta" }