I'm a bit confused about one sentence in the timeline of this competition:\n\"Model upload & first stage deadline. This is the last day you may upload your model to be eligible for a prize.\"
\n\nwhat is meant with the \"model upload\"? in what format should the model be?\nhow will it be checked that the final submission was made with the same model and is that a requirement?
", "messages": [ { "id": "177756", "postDate": "04/25/2017 20:19:39", "content": "I'm a bit confused about one sentence in the timeline of this competition:\n\"Model upload & first stage deadline. This is the last day you may upload your model to be eligible for a prize.\"
\n\nwhat is meant with the \"model upload\"? in what format should the model be?\nhow will it be checked that the final submission was made with the same model and is that a requirement?
", "rawMarkdown": "I'm a bit confused about one sentence in the timeline of this competition:\n\"Model upload & first stage deadline. This is the last day you may upload your model to be eligible for a prize.\"\n\nwhat is meant with the \"model upload\"? in what format should the model be?\nhow will it be checked that the final submission was made with the same model and is that a requirement?", "votes": null }, { "id": "177790", "postDate": "04/25/2017 22:58:37", "content": "The model you upload is everything required to produce your LB score. It will include all scripts and weights trained. Kaggle admins will validate your approach if you happen to win a prize. As with other competitions, you have to upload a model to be considered for a medal, so it's in your interest to make your implementation, as reproducible as possible.
", "rawMarkdown": "The model you upload is everything required to produce your LB score. It will include all scripts and weights trained. Kaggle admins will validate your approach if you happen to win a prize. As with other competitions, you have to upload a model to be considered for a medal, so it's in your interest to make your implementation, as reproducible as possible.", "votes": null }, { "id": "177911", "postDate": "04/26/2017 06:20:08", "content": "Thanks, but upload where? And I guess we're not limited to one model but can use an ensemble etc. so i can imagine that it will be few gigs for some people.
", "rawMarkdown": "Thanks, but upload where? And I guess we're not limited to one model but can use an ensemble etc. so i can imagine that it will be few gigs for some people.", "votes": null }, { "id": "178451", "postDate": "04/27/2017 20:45:15", "content": "For a previous competition this model upload area was under More > Team. But it didn't appear until a few days before the deadline.
\n\nYou are not limited--you can ensemble. It's just that you have to upload the code to reproduce the submissions that you choose at the time of deadline.
\n\nThis is set up very much like the Nature Conservancy Fisheries competition and there were many questions about why it was set up with two test stages. You can think of this setup as a, let's say, 50/50 split of test data. Except, they don't want you to see the 50% of test data used for the Private Leaderboard. This is in contrast to other competitions where they may say the Public Leaderboard is calculated with 50% of the Test set. But, you can still look at all 100% of the Test set, which could in turn be used to improve your models.
", "rawMarkdown": "For a previous competition this model upload area was under More > Team. But it didn't appear until a few days before the deadline. \n\nYou are not limited--you can ensemble. It's just that you have to upload the code to reproduce the submissions that you choose at the time of deadline. \n\nThis is set up very much like the Nature Conservancy Fisheries competition and there were many questions about why it was set up with two test stages. You can think of this setup as a, let's say, 50/50 split of test data. Except, they don't want you to see the 50% of test data used for the Private Leaderboard. This is in contrast to other competitions where they may say the Public Leaderboard is calculated with 50% of the Test set. But, you can still look at all 100% of the Test set, which could in turn be used to improve your models.", "votes": null }, { "id": "178543", "postDate": "04/28/2017 06:19:47", "content": "Thank you, that was the answer I was looking for ;) until now I never saw such a competition and couldn't see any link to an upload area
", "rawMarkdown": "Thank you, that was the answer I was looking for ;) until now I never saw such a competition and couldn't see any link to an upload area", "votes": null } ], "comments": [ { "id": 177790, "author_name": "craigglastonbury", "author_url": "", "post_date": "04/25/2017 22:58:37", "content": "The model you upload is everything required to produce your LB score. It will include all scripts and weights trained. Kaggle admins will validate your approach if you happen to win a prize. As with other competitions, you have to upload a model to be considered for a medal, so it's in your interest to make your implementation, as reproducible as possible.
", "votes": null, "replies": [ { "id": 177911, "author_name": "steelrose", "author_url": "", "post_date": "04/26/2017 06:20:08", "content": "Thanks, but upload where? And I guess we're not limited to one model but can use an ensemble etc. so i can imagine that it will be few gigs for some people.
", "votes": null, "replies": [] }, { "id": 178451, "author_name": "mccroden", "author_url": "", "post_date": "04/27/2017 20:45:15", "content": "For a previous competition this model upload area was under More > Team. But it didn't appear until a few days before the deadline.
\n\nYou are not limited--you can ensemble. It's just that you have to upload the code to reproduce the submissions that you choose at the time of deadline.
\n\nThis is set up very much like the Nature Conservancy Fisheries competition and there were many questions about why it was set up with two test stages. You can think of this setup as a, let's say, 50/50 split of test data. Except, they don't want you to see the 50% of test data used for the Private Leaderboard. This is in contrast to other competitions where they may say the Public Leaderboard is calculated with 50% of the Test set. But, you can still look at all 100% of the Test set, which could in turn be used to improve your models.
", "votes": null, "replies": [] }, { "id": 178543, "author_name": "steelrose", "author_url": "", "post_date": "04/28/2017 06:19:47", "content": "Thank you, that was the answer I was looking for ;) until now I never saw such a competition and couldn't see any link to an upload area
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "177756": "I'm a bit confused about one sentence in the timeline of this competition:\n\"Model upload & first stage deadline. This is the last day you may upload your model to be eligible for a prize.\"\n\nwhat is meant with the \"model upload\"? in what format should the model be?\nhow will it be checked that the final submission was made with the same model and is that a requirement?", "177790": "The model you upload is everything required to produce your LB score. It will include all scripts and weights trained. Kaggle admins will validate your approach if you happen to win a prize. As with other competitions, you have to upload a model to be considered for a medal, so it's in your interest to make your implementation, as reproducible as possible.", "177911": "Thanks, but upload where? And I guess we're not limited to one model but can use an ensemble etc. so i can imagine that it will be few gigs for some people.", "178451": "For a previous competition this model upload area was under More > Team. But it didn't appear until a few days before the deadline. \n\nYou are not limited--you can ensemble. It's just that you have to upload the code to reproduce the submissions that you choose at the time of deadline. \n\nThis is set up very much like the Nature Conservancy Fisheries competition and there were many questions about why it was set up with two test stages. You can think of this setup as a, let's say, 50/50 split of test data. Except, they don't want you to see the 50% of test data used for the Private Leaderboard. This is in contrast to other competitions where they may say the Public Leaderboard is calculated with 50% of the Test set. But, you can still look at all 100% of the Test set, which could in turn be used to improve your models.", "178543": "Thank you, that was the answer I was looking for ;) until now I never saw such a competition and couldn't see any link to an upload area" }, "source": "meta" }