Hi,
\n\nI'm confused about the model upload requirement.
\n\nI don't understand how the model could be verified unless the environment this model runs in is specified.
\n\nFor example, let's say that I really think that FORTRAN 66 is the right choice for my work. If I upload my model from my FORTRAN 66 program it would look like a bunch of printed FORTRAN 66 arrays.
\n\nOr, should I upload my FORTRAN 66 program? In that case won't there have to be specifications about the environment since my choice could be the best FORTRAN 66 environment, ie, a VAX.
\n\nSo, what is the environment that the model has to run in?
\n\nThanks.\nbruce
", "messages": [ { "id": 376968, "postDate": "2018-08-28T12:00:11.387Z", "content": "[opinion from a longtime kaggler, not from the official team]
\n\nUploading the model is to show that your work was done based on stage 1 and not significantly changed once the stage 2 data is out. You are uploading your source code. It does not have to run in any specific environment. To get a final ranking for the competition, you must upload a model by the deadline (and submit an entry in both stage 1 and stage 2).
\n\nI don't know the details, but I would presume that if you are in the top ten for this contest (since there are ten money prizes), you will have to show that your final model matches your stage 1 model. In past contests I have seen them have all grey-scale for stage 1 and then add some RGB for stage 2. I presume minor tweaks to the model to compensate for such things is allowed, but you are not allowed to make significant changes to your model or manually retrain your hyperparameters based on the stage 2 data. Only Kaggle can tell you how much of a change you can make to your model for stage 2, but I would plan on making no changes, since that seems to be what a strict reading of the rules say.
\n\nGood luck!
", "rawMarkdown": "[opinion from a longtime kaggler, not from the official team]\n\nUploading the model is to show that your work was done based on stage 1 and not significantly changed once the stage 2 data is out. You are uploading your source code. It does not have to run in any specific environment. To get a final ranking for the competition, you must upload a model by the deadline (and submit an entry in both stage 1 and stage 2).\n\nI don't know the details, but I would presume that if you are in the top ten for this contest (since there are ten money prizes), you will have to show that your final model matches your stage 1 model. In past contests I have seen them have all grey-scale for stage 1 and then add some RGB for stage 2. I presume minor tweaks to the model to compensate for such things is allowed, but you are not allowed to make significant changes to your model or manually retrain your hyperparameters based on the stage 2 data. Only Kaggle can tell you how much of a change you can make to your model for stage 2, but I would plan on making no changes, since that seems to be what a strict reading of the rules say.\n\nGood luck!", "votes": 1, "replies": [ { "id": 377155, "postDate": "2018-08-28T16:49:23.280Z", "content": "From the \"official team,\" I could not have said it better myself ^ (Thank you)
\n\nThe Stage 1 model upload should be your source code which satisfies the intent specified in the rules and Timeline page. Stage 2 winners will be expected to provide more extensive documentation of their solutions, and models are not permitted to change between Stage 1 and Stage 2.
", "rawMarkdown": "From the \"official team,\" I could not have said it better myself ^ (Thank you)\n\nThe Stage 1 model upload should be your source code which satisfies the intent specified in the rules and [Timeline page][2]. Stage 2 winners will be expected to provide more [extensive documentation][1] of their solutions, and models are not permitted to change between Stage 1 and Stage 2.\n\n\n[1]: https://www.kaggle.com/WinningModelDocumentationGuidelines\n[2]: https://www.kaggle.com/c/rsna-pneumonia-detection-challenge#Timeline" } ] }, { "id": 376949, "postDate": "2018-08-28T11:13:07.440Z", "content": "Hi,
\n\nI'm confused about the model upload requirement.
\n\nI don't understand how the model could be verified unless the environment this model runs in is specified.
\n\nFor example, let's say that I really think that FORTRAN 66 is the right choice for my work. If I upload my model from my FORTRAN 66 program it would look like a bunch of printed FORTRAN 66 arrays.
\n\nOr, should I upload my FORTRAN 66 program? In that case won't there have to be specifications about the environment since my choice could be the best FORTRAN 66 environment, ie, a VAX.
\n\nSo, what is the environment that the model has to run in?
\n\nThanks.\nbruce
", "rawMarkdown": "Hi,\n\nI'm confused about the model upload requirement.\n\nI don't understand how the model could be verified unless the environment this model runs in is specified.\n\nFor example, let's say that I really think that FORTRAN 66 is the right choice for my work. If I upload my model from my FORTRAN 66 program it would look like a bunch of printed FORTRAN 66 arrays. \n\nOr, should I upload my FORTRAN 66 program? In that case won't there have to be specifications about the environment since my choice could be the best FORTRAN 66 environment, ie, a VAX.\n\nSo, what is the environment that the model has to run in?\n\nThanks.\nbruce", "votes": 1 }, { "id": 377162, "postDate": "2018-08-28T16:57:30.123Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 376968, "author_name": "quadcore/Richard Epstein", "author_url": "", "post_date": "2018-08-28T12:00:11.387000", "content": "[opinion from a longtime kaggler, not from the official team]
\n\nUploading the model is to show that your work was done based on stage 1 and not significantly changed once the stage 2 data is out. You are uploading your source code. It does not have to run in any specific environment. To get a final ranking for the competition, you must upload a model by the deadline (and submit an entry in both stage 1 and stage 2).
\n\nI don't know the details, but I would presume that if you are in the top ten for this contest (since there are ten money prizes), you will have to show that your final model matches your stage 1 model. In past contests I have seen them have all grey-scale for stage 1 and then add some RGB for stage 2. I presume minor tweaks to the model to compensate for such things is allowed, but you are not allowed to make significant changes to your model or manually retrain your hyperparameters based on the stage 2 data. Only Kaggle can tell you how much of a change you can make to your model for stage 2, but I would plan on making no changes, since that seems to be what a strict reading of the rules say.
\n\nGood luck!
", "votes": 1, "replies": [ { "id": 377155, "author_name": "Julia Elliott", "author_url": "", "post_date": "2018-08-28T16:49:23.280000", "content": "From the \"official team,\" I could not have said it better myself ^ (Thank you)
\n\nThe Stage 1 model upload should be your source code which satisfies the intent specified in the rules and Timeline page. Stage 2 winners will be expected to provide more extensive documentation of their solutions, and models are not permitted to change between Stage 1 and Stage 2.
", "votes": 0, "replies": [] } ] }, { "id": 377162, "author_name": "", "author_url": "", "post_date": "2018-08-28T16:57:30.123000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "376968": "[opinion from a longtime kaggler, not from the official team]\n\nUploading the model is to show that your work was done based on stage 1 and not significantly changed once the stage 2 data is out. You are uploading your source code. It does not have to run in any specific environment. To get a final ranking for the competition, you must upload a model by the deadline (and submit an entry in both stage 1 and stage 2).\n\nI don't know the details, but I would presume that if you are in the top ten for this contest (since there are ten money prizes), you will have to show that your final model matches your stage 1 model. In past contests I have seen them have all grey-scale for stage 1 and then add some RGB for stage 2. I presume minor tweaks to the model to compensate for such things is allowed, but you are not allowed to make significant changes to your model or manually retrain your hyperparameters based on the stage 2 data. Only Kaggle can tell you how much of a change you can make to your model for stage 2, but I would plan on making no changes, since that seems to be what a strict reading of the rules say.\n\nGood luck!", "376949": "Hi,\n\nI'm confused about the model upload requirement.\n\nI don't understand how the model could be verified unless the environment this model runs in is specified.\n\nFor example, let's say that I really think that FORTRAN 66 is the right choice for my work. If I upload my model from my FORTRAN 66 program it would look like a bunch of printed FORTRAN 66 arrays. \n\nOr, should I upload my FORTRAN 66 program? In that case won't there have to be specifications about the environment since my choice could be the best FORTRAN 66 environment, ie, a VAX.\n\nSo, what is the environment that the model has to run in?\n\nThanks.\nbruce", "377162": "" } }