The Earthquake is really shake in the leaderboard.\nMany people unselected the true submission, me too. \nI lost my silver medal. I screwed up.\nJust trust your local CV, this is what I learn from this game. Congrats to winners.
", "messages": [ { "id": "542690", "postDate": "06/04/2019 03:34:20", "content": "The Earthquake is really shake in the leaderboard.\nMany people unselected the true submission, me too. \nI lost my silver medal. I screwed up.\nJust trust your local CV, this is what I learn from this game. Congrats to winners.
", "rawMarkdown": "The Earthquake is really shake in the leaderboard.\nMany people unselected the true submission, me too. \nI lost my silver medal. I screwed up.\n**Just trust your local CV**, this is what I learn from this game. Congrats to winners.", "votes": null }, { "id": "542694", "postDate": "06/04/2019 03:36:40", "content": "It's the opposite to me. DON'T TRUST YOUR CV TOO MUCH, but try to exploit the data leak. You can simply win by that. And even more than that, I bet if no CV can guarantee for a good private LB score.
", "rawMarkdown": "It's the opposite to me. DON'T TRUST YOUR CV TOO MUCH, but try to exploit the data leak. You can simply win by that. And even more than that, I bet if no CV can guarantee for a good private LB score.", "votes": null }, { "id": "542740", "postDate": "06/04/2019 04:18:09", "content": "I always say: trust your local validation strategy. (But only if your validation is robust and built correctly).
", "rawMarkdown": "I always say: trust your local validation strategy. (But only if your validation is robust and built correctly).", "votes": null }, { "id": "542744", "postDate": "06/04/2019 04:23:49", "content": "i think i somewhat agree with @carlospk who have very huge jump in lb and gave him gold medals 6th position. The separate scaling of feature for training and testing is very important , so he can manage to avoid mean shifting effect in private test.
", "rawMarkdown": "i think i somewhat agree with @carlospk who have very huge jump in lb and gave him gold medals 6th position. The separate scaling of feature for training and testing is very important , so he can manage to avoid mean shifting effect in private test.", "votes": null }, { "id": "542747", "postDate": "06/04/2019 04:26:53", "content": "I totally disagree 100% with you. Did you use any information from the image of test EQ? Sorry, but you even admitted you used that information in another topic. If somebody is not aware of that information, I bet no CV can do well on private LB with only the training data. If you disagree, would you mind explain what is your CV using only train data and pretending not knowing anything about the private test set?
", "rawMarkdown": "I totally disagree 100% with you. Did you use any information from the image of test EQ? Sorry, but you even admitted you used that information in another topic. If somebody is not aware of that information, I bet no CV can do well on private LB with only the training data. If you disagree, would you mind explain what is your CV using only train data and pretending not knowing anything about the private test set?", "votes": null } ], "comments": [ { "id": 542694, "author_name": "khahuras", "author_url": "", "post_date": "06/04/2019 03:36:40", "content": "It's the opposite to me. DON'T TRUST YOUR CV TOO MUCH, but try to exploit the data leak. You can simply win by that. And even more than that, I bet if no CV can guarantee for a good private LB score.
", "votes": null, "replies": [] }, { "id": 542740, "author_name": "titericz", "author_url": "", "post_date": "06/04/2019 04:18:09", "content": "I always say: trust your local validation strategy. (But only if your validation is robust and built correctly).
", "votes": null, "replies": [ { "id": 542747, "author_name": "khahuras", "author_url": "", "post_date": "06/04/2019 04:26:53", "content": "I totally disagree 100% with you. Did you use any information from the image of test EQ? Sorry, but you even admitted you used that information in another topic. If somebody is not aware of that information, I bet no CV can do well on private LB with only the training data. If you disagree, would you mind explain what is your CV using only train data and pretending not knowing anything about the private test set?
", "votes": null, "replies": [] } ] }, { "id": 542744, "author_name": "arisukma", "author_url": "", "post_date": "06/04/2019 04:23:49", "content": "i think i somewhat agree with @carlospk who have very huge jump in lb and gave him gold medals 6th position. The separate scaling of feature for training and testing is very important , so he can manage to avoid mean shifting effect in private test.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "542690": "The Earthquake is really shake in the leaderboard.\nMany people unselected the true submission, me too. \nI lost my silver medal. I screwed up.\n**Just trust your local CV**, this is what I learn from this game. Congrats to winners.", "542694": "It's the opposite to me. DON'T TRUST YOUR CV TOO MUCH, but try to exploit the data leak. You can simply win by that. And even more than that, I bet if no CV can guarantee for a good private LB score.", "542740": "I always say: trust your local validation strategy. (But only if your validation is robust and built correctly).", "542744": "i think i somewhat agree with @carlospk who have very huge jump in lb and gave him gold medals 6th position. The separate scaling of feature for training and testing is very important , so he can manage to avoid mean shifting effect in private test.", "542747": "I totally disagree 100% with you. Did you use any information from the image of test EQ? Sorry, but you even admitted you used that information in another topic. If somebody is not aware of that information, I bet no CV can do well on private LB with only the training data. If you disagree, would you mind explain what is your CV using only train data and pretending not knowing anything about the private test set?" }, "source": "meta" }