Congratulations to the winners teams: w, MA and slonoschildpad \nYou guys did great in this competition! Very consistent trajectory and Public/Private LB scores. As I always say: - Trust your local CV.
\n\nAlso congrats to top 13 golden medal teams, but a special congrats to: Lingzhi, BVoskresenskii, mykper, Jordan Meyer, AndreyGo, antklen and Louis.T, who made a solo golden medal. This is the hardest pre-requisite to become a GM.
\n\nAlso I would like to thanks Kaggle for hosting and Favorita for sponsoring this great competition.
", "messages": [ { "id": "269038", "postDate": "01/16/2018 04:54:32", "content": "Congratulations to the winners teams: w, MA and slonoschildpad \nYou guys did great in this competition! Very consistent trajectory and Public/Private LB scores. As I always say: - Trust your local CV.
\n\nAlso congrats to top 13 golden medal teams, but a special congrats to: Lingzhi, BVoskresenskii, mykper, Jordan Meyer, AndreyGo, antklen and Louis.T, who made a solo golden medal. This is the hardest pre-requisite to become a GM.
\n\nAlso I would like to thanks Kaggle for hosting and Favorita for sponsoring this great competition.
", "rawMarkdown": "Congratulations to the winners teams: w, MA and slonoschildpad \nYou guys did great in this competition! Very consistent trajectory and Public/Private LB scores. As I always say: - Trust your local CV.\n\nAlso congrats to top 13 golden medal teams, but a special congrats to: Lingzhi, BVoskresenskii, mykper, Jordan Meyer, AndreyGo, antklen and Louis.T, who made a solo golden medal. This is the hardest pre-requisite to become a GM.\n\nAlso I would like to thanks Kaggle for hosting and Favorita for sponsoring this great competition.", "votes": null }, { "id": "269052", "postDate": "01/16/2018 05:29:40", "content": "\n\n\nAs I always say: - Trust your local CV.
\n
If you trust your local CV in this competition, your model will be biased, correct?
", "rawMarkdown": "> As I always say: - Trust your local CV.\n\nIf you trust your local CV in this competition, your model will be biased, correct?", "votes": null }, { "id": "269083", "postDate": "01/16/2018 07:00:44", "content": "And why you local CV should be best, that LB?
", "rawMarkdown": "And why you local CV should be best, that LB?", "votes": null }, { "id": "269087", "postDate": "01/16/2018 07:07:51", "content": "Biased due onpromotion?
\nIn this competition we spent a few weeks searching for a good way to validate our models. We changed the validation strategy a couple of times. By the end it performed consistently improving locally and LB.
And what is you best validation strategy?
", "rawMarkdown": "And what is you best validation strategy?", "votes": null }, { "id": "269096", "postDate": "01/16/2018 07:29:54", "content": "I'm sure my teammate will write about it tomorrow. It's not the best, but it worked well for us. We choose two validation periods, one from 2017-07-16 to 2017-07-31 and other from 2017-08-01 to 2017-08-15 mimicking the testset. Not only the periods, but how to build the features is also important here.
", "rawMarkdown": "I'm sure my teammate will write about it tomorrow. It's not the best, but it worked well for us. We choose two validation periods, one from 2017-07-16 to 2017-07-31 and other from 2017-08-01 to 2017-08-15 mimicking the testset. Not only the periods, but how to build the features is also important here.", "votes": null }, { "id": "269116", "postDate": "01/16/2018 08:21:37", "content": "Thanks very much!
", "rawMarkdown": "Thanks very much!", "votes": null }, { "id": "269135", "postDate": "01/16/2018 09:33:55", "content": "Thank you! Congrats to you too! =)
", "rawMarkdown": "Thank you! Congrats to you too! =)", "votes": null }, { "id": "269228", "postDate": "01/16/2018 13:52:24", "content": "Because public LB gives you no information about how your models can predict 6 to 16 days in the future, while local CV does. Unless you use the public LB with at least 5 day lag.
", "rawMarkdown": "Because public LB gives you no information about how your models can predict 6 to 16 days in the future, while local CV does. Unless you use the public LB with at least 5 day lag.", "votes": null } ], "comments": [ { "id": 269052, "author_name": "seanjv", "author_url": "", "post_date": "01/16/2018 05:29:40", "content": "\n\n\nAs I always say: - Trust your local CV.
\n
If you trust your local CV in this competition, your model will be biased, correct?
", "votes": null, "replies": [ { "id": 269087, "author_name": "titericz", "author_url": "", "post_date": "01/16/2018 07:07:51", "content": "Biased due onpromotion?
\nIn this competition we spent a few weeks searching for a good way to validate our models. We changed the validation strategy a couple of times. By the end it performed consistently improving locally and LB.
And what is you best validation strategy?
", "votes": null, "replies": [] }, { "id": 269096, "author_name": "titericz", "author_url": "", "post_date": "01/16/2018 07:29:54", "content": "I'm sure my teammate will write about it tomorrow. It's not the best, but it worked well for us. We choose two validation periods, one from 2017-07-16 to 2017-07-31 and other from 2017-08-01 to 2017-08-15 mimicking the testset. Not only the periods, but how to build the features is also important here.
", "votes": null, "replies": [] } ] }, { "id": 269083, "author_name": "zavodrobotov", "author_url": "", "post_date": "01/16/2018 07:00:44", "content": "And why you local CV should be best, that LB?
", "votes": null, "replies": [ { "id": 269228, "author_name": "cpmpml", "author_url": "", "post_date": "01/16/2018 13:52:24", "content": "Because public LB gives you no information about how your models can predict 6 to 16 days in the future, while local CV does. Unless you use the public LB with at least 5 day lag.
", "votes": null, "replies": [] } ] }, { "id": 269116, "author_name": "luciusyu", "author_url": "", "post_date": "01/16/2018 08:21:37", "content": "Thanks very much!
", "votes": null, "replies": [] }, { "id": 269135, "author_name": "bvoskresenskii", "author_url": "", "post_date": "01/16/2018 09:33:55", "content": "Thank you! Congrats to you too! =)
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "269038": "Congratulations to the winners teams: w, MA and slonoschildpad \nYou guys did great in this competition! Very consistent trajectory and Public/Private LB scores. As I always say: - Trust your local CV.\n\nAlso congrats to top 13 golden medal teams, but a special congrats to: Lingzhi, BVoskresenskii, mykper, Jordan Meyer, AndreyGo, antklen and Louis.T, who made a solo golden medal. This is the hardest pre-requisite to become a GM.\n\nAlso I would like to thanks Kaggle for hosting and Favorita for sponsoring this great competition.", "269052": "> As I always say: - Trust your local CV.\n\nIf you trust your local CV in this competition, your model will be biased, correct?", "269083": "And why you local CV should be best, that LB?", "269087": "Biased due onpromotion? \nIn this competition we spent a few weeks searching for a good way to validate our models. We changed the validation strategy a couple of times. By the end it performed consistently improving locally and LB.", "269092": "And what is you best validation strategy?", "269096": "I'm sure my teammate will write about it tomorrow. It's not the best, but it worked well for us. We choose two validation periods, one from 2017-07-16 to 2017-07-31 and other from 2017-08-01 to 2017-08-15 mimicking the testset. Not only the periods, but how to build the features is also important here.", "269116": "Thanks very much!", "269135": "Thank you! Congrats to you too! =)", "269228": "Because public LB gives you no information about how your models can predict 6 to 16 days in the future, while local CV does. Unless you use the public LB with at least 5 day lag." }, "source": "meta" }