I have an access to brand new gaming computer which has got Nvidia GTX 3050 Ti GPU. As I am realitively new to kaggle, after reading coding requirements which puts some restrictions about pretrained models, I am wondering how I can use this machine to train a model for hours to get higher score.
\nAccording to my thinking, I will download all the data, build a model, train it for hours/days and after optimizing I will upload the saved model to generate a submission file.
\nIn this case I have questions:
\nShould I train a model in my machine upto 12 hour as told in code requirments or unlimited ?
\nThank you very much for replying!
\nGood luck to everyone!
I have an access to brand new gaming computer which has got Nvidia GTX 3050 Ti GPU. As I am realitively new to kaggle, after reading coding requirements which puts some restrictions about pretrained models, I am wondering how I can use this machine to train a model for hours to get higher score.
\nAccording to my thinking, I will download all the data, build a model, train it for hours/days and after optimizing I will upload the saved model to generate a submission file.
\nIn this case I have questions:
\nShould I train a model in my machine upto 12 hour as told in code requirments or unlimited ?
\nThank you very much for replying!
\nGood luck to everyone!
The coding requirements are only for inference. You can train your model wherever you want and for however long you want. It is also ok to upload your weights as private datasets, it is not against the rules. In a notebook other than your training notebook, you can generate predictions using the weights. The running time for this notebook has to be under 12 hours.
", "rawMarkdown": "The coding requirements are only for inference. You can train your model wherever you want and for however long you want. It is also ok to upload your weights as private datasets, it is not against the rules. In a notebook other than your training notebook, you can generate predictions using the weights. The running time for this notebook has to be under 12 hours.", "votes": null }, { "id": "1514387", "postDate": "09/16/2021 03:22:36", "content": "Thank you bro for the reply.
", "rawMarkdown": "Thank you bro for the reply.", "votes": null }, { "id": "1514465", "postDate": "09/16/2021 05:30:02", "content": "You can do. it is always easy for longer training but still I would say you must do it in kaggle's interface because it has perfect environment setup for you and dataset pre loaded. Secondly, execution time for competition is 9-12hr you local machine may do it earlier or late and fail the notebook's comp.
", "rawMarkdown": "You can do. it is always easy for longer training but still I would say you must do it in kaggle's interface because it has perfect environment setup for you and dataset pre loaded. Secondly, execution time for competition is 9-12hr you local machine may do it earlier or late and fail the notebook's comp.", "votes": null } ], "comments": [ { "id": 1514266, "author_name": "novice03", "author_url": "", "post_date": "09/15/2021 21:16:29", "content": "The coding requirements are only for inference. You can train your model wherever you want and for however long you want. It is also ok to upload your weights as private datasets, it is not against the rules. In a notebook other than your training notebook, you can generate predictions using the weights. The running time for this notebook has to be under 12 hours.
", "votes": null, "replies": [ { "id": 1514387, "author_name": "abbosjon", "author_url": "", "post_date": "09/16/2021 03:22:36", "content": "Thank you bro for the reply.
", "votes": null, "replies": [] } ] }, { "id": 1514465, "author_name": "eryash15", "author_url": "", "post_date": "09/16/2021 05:30:02", "content": "You can do. it is always easy for longer training but still I would say you must do it in kaggle's interface because it has perfect environment setup for you and dataset pre loaded. Secondly, execution time for competition is 9-12hr you local machine may do it earlier or late and fail the notebook's comp.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1514246": "I have an access to brand new gaming computer which has got Nvidia GTX 3050 Ti GPU. As I am realitively new to kaggle, after reading **coding requirements** which puts some restrictions about pretrained models, I am wondering how I can use this machine to train a model for hours to get higher score.\n\nAccording to my thinking, I will download all the data, build a model, train it for hours/days and after optimizing I will upload the saved model to generate a submission file.\n\nIn this case I have questions:\n\nShould I train a model in my machine upto 12 hour as told in code requirments or unlimited ?\n1. If I just upload my pre-trained model, isn't it considered as a private data?\n2. or should I also publicly provide it on kaggle datasets?\n3. if my hyposesis is wrong, is there any task I can use my machine for getting higher ranks in a leaderboard?\n\nThank you very much for replying!\nGood luck to everyone!", "1514266": "The coding requirements are only for inference. You can train your model wherever you want and for however long you want. It is also ok to upload your weights as private datasets, it is not against the rules. In a notebook other than your training notebook, you can generate predictions using the weights. The running time for this notebook has to be under 12 hours.", "1514387": "Thank you bro for the reply.", "1514465": "You can do. it is always easy for longer training but still I would say you must do it in kaggle's interface because it has perfect environment setup for you and dataset pre loaded. Secondly, execution time for competition is 9-12hr you local machine may do it earlier or late and fail the notebook's comp." }, "source": "meta" }