A strong ensemble kernel often appears immediately before the deadline. You should not submit the submission file as is. You should change the id=0's predicted label to 0 or 1. If id=0 is included in private LB, either score must beat the original. This gives you a higher rank. I found this just after the deadline :(
", "messages": [ { "id": "325551", "postDate": "05/08/2018 14:39:04", "content": "A strong ensemble kernel often appears immediately before the deadline. You should not submit the submission file as is. You should change the id=0's predicted label to 0 or 1. If id=0 is included in private LB, either score must beat the original. This gives you a higher rank. I found this just after the deadline :(
", "rawMarkdown": "A strong ensemble kernel often appears immediately before the deadline. You should not submit the submission file as is. You should change the id=0's predicted label to 0 or 1. If id=0 is included in private LB, either score must beat the original. This gives you a higher rank. I found this just after the deadline :(", "votes": null }, { "id": "325565", "postDate": "05/08/2018 14:50:58", "content": "Smart!
", "rawMarkdown": "Smart!", "votes": null }, { "id": "325940", "postDate": "05/09/2018 05:04:19", "content": "Needs to care the case of traps of random noise in private test data suggested by @pocket. You will loose both final submissions otherwise.\nhttps://www.kaggle.com/c/talkingdata-adtracking-fraud-detection/discussion/56182#324655
", "rawMarkdown": "Needs to care the case of traps of random noise in private test data suggested by @pocket. You will loose both final submissions otherwise.\nhttps://www.kaggle.com/c/talkingdata-adtracking-fraud-detection/discussion/56182#324655", "votes": null }, { "id": "326014", "postDate": "05/09/2018 06:54:50", "content": "Good one!
", "rawMarkdown": "Good one!", "votes": null }, { "id": "326896", "postDate": "05/10/2018 13:27:31", "content": "Thank you for your important advice. Yes, we have to take care of overfitting to the public LB not only when it's fake.
", "rawMarkdown": "Thank you for your important advice. Yes, we have to take care of overfitting to the public LB not only when it's fake.", "votes": null }, { "id": "327177", "postDate": "05/11/2018 00:57:52", "content": "This is a kind of hand labeling of test data, which is explicitly prohibited by the Rule of this competition. The Rule says:
\n\n\n\n\nSubmissions may not use or incorporate information from hand labeling or human prediction of the validation dataset or test data records.
\n
In addition to that, the umeshsati54's kernel in this competition was using a dataset of deleted kernel. That makes your submission unreporoducible by using the original datasets. I think this also can be against for something.
", "rawMarkdown": "This is a kind of hand labeling of test data, which is explicitly prohibited by the Rule of this competition. The Rule says:\n\n> Submissions may not use or incorporate information from hand labeling or human prediction of the validation dataset or test data records.\n\nIn addition to that, the umeshsati54's kernel in this competition was using a dataset of deleted kernel. That makes your submission unreporoducible by using the original datasets. I think this also can be against for something.", "votes": null }, { "id": "327427", "postDate": "05/11/2018 14:35:00", "content": "In my understanding, 0 or 1 prediction is not a hand labeling but a strategy. this technique is broadly used in NCAA competition to predict the final match of the tournament for sure.
", "rawMarkdown": "In my understanding, 0 or 1 prediction is not a hand labeling but a strategy. this technique is broadly used in NCAA competition to predict the final match of the tournament for sure.", "votes": null } ], "comments": [ { "id": 325565, "author_name": "cpmpml", "author_url": "", "post_date": "05/08/2018 14:50:58", "content": "Smart!
", "votes": null, "replies": [] }, { "id": 325940, "author_name": "fujihiro", "author_url": "", "post_date": "05/09/2018 05:04:19", "content": "Needs to care the case of traps of random noise in private test data suggested by @pocket. You will loose both final submissions otherwise.\nhttps://www.kaggle.com/c/talkingdata-adtracking-fraud-detection/discussion/56182#324655
", "votes": null, "replies": [ { "id": 326896, "author_name": "osciiart", "author_url": "", "post_date": "05/10/2018 13:27:31", "content": "Thank you for your important advice. Yes, we have to take care of overfitting to the public LB not only when it's fake.
", "votes": null, "replies": [] } ] }, { "id": 326014, "author_name": "ericbenhamou", "author_url": "", "post_date": "05/09/2018 06:54:50", "content": "Good one!
", "votes": null, "replies": [] }, { "id": 327177, "author_name": "fujihiro", "author_url": "", "post_date": "05/11/2018 00:57:52", "content": "This is a kind of hand labeling of test data, which is explicitly prohibited by the Rule of this competition. The Rule says:
\n\n\n\n\nSubmissions may not use or incorporate information from hand labeling or human prediction of the validation dataset or test data records.
\n
In addition to that, the umeshsati54's kernel in this competition was using a dataset of deleted kernel. That makes your submission unreporoducible by using the original datasets. I think this also can be against for something.
", "votes": null, "replies": [ { "id": 327427, "author_name": "osciiart", "author_url": "", "post_date": "05/11/2018 14:35:00", "content": "In my understanding, 0 or 1 prediction is not a hand labeling but a strategy. this technique is broadly used in NCAA competition to predict the final match of the tournament for sure.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "325551": "A strong ensemble kernel often appears immediately before the deadline. You should not submit the submission file as is. You should change the id=0's predicted label to 0 or 1. If id=0 is included in private LB, either score must beat the original. This gives you a higher rank. I found this just after the deadline :(", "325565": "Smart!", "325940": "Needs to care the case of traps of random noise in private test data suggested by @pocket. You will loose both final submissions otherwise.\nhttps://www.kaggle.com/c/talkingdata-adtracking-fraud-detection/discussion/56182#324655", "326014": "Good one!", "326896": "Thank you for your important advice. Yes, we have to take care of overfitting to the public LB not only when it's fake.", "327177": "This is a kind of hand labeling of test data, which is explicitly prohibited by the Rule of this competition. The Rule says:\n\n> Submissions may not use or incorporate information from hand labeling or human prediction of the validation dataset or test data records.\n\nIn addition to that, the umeshsati54's kernel in this competition was using a dataset of deleted kernel. That makes your submission unreporoducible by using the original datasets. I think this also can be against for something.", "327427": "In my understanding, 0 or 1 prediction is not a hand labeling but a strategy. this technique is broadly used in NCAA competition to predict the final match of the tournament for sure." }, "source": "meta" }