Hey!
\nIf I understand correctly, in a code competition I can still train and experiment locally on my own machine. I can then upload trained models and perform inference via a kaggle kernel?
\nMeaning, it's okay if I train the models elsewhere, not on Kaggle? My kernel can just do inference?
\nBTW, the time limit is 9 hrs for the kernel. But how do we know how much data will be made available to our model in the subsequent phases of the competition?
", "messages": [ { "id": "1795075", "postDate": "05/19/2022 11:50:34", "content": "Hey!
\nIf I understand correctly, in a code competition I can still train and experiment locally on my own machine. I can then upload trained models and perform inference via a kaggle kernel?
\nMeaning, it's okay if I train the models elsewhere, not on Kaggle? My kernel can just do inference?
\nBTW, the time limit is 9 hrs for the kernel. But how do we know how much data will be made available to our model in the subsequent phases of the competition?
", "rawMarkdown": "Hey!\n\nIf I understand correctly, in a code competition I can still train and experiment locally on my own machine. I can then upload trained models and perform inference via a kaggle kernel?\n\nMeaning, it's okay if I train the models elsewhere, not on Kaggle? My kernel can just do inference?\n\nBTW, the time limit is 9 hrs for the kernel. But how do we know how much data will be made available to our model in the subsequent phases of the competition?", "votes": null }, { "id": "1795089", "postDate": "05/19/2022 11:59:43", "content": "yes, you can train your model else where. There will be about 20000 notebooks for test phase (check data tab).
", "rawMarkdown": "yes, you can train your model else where. There will be about 20000 notebooks for test phase (check data tab).", "votes": null }, { "id": "1795213", "postDate": "05/19/2022 14:35:25", "content": "\n\ndo we know how much data will be made available to our model in the subsequent phases of the competition?
\n
The time limit remains the same for both stages.
", "rawMarkdown": "> do we know how much data will be made available to our model in the subsequent phases of the competition?\n\nThe time limit remains the same for both stages.", "votes": null }, { "id": "1795287", "postDate": "05/19/2022 15:53:25", "content": "Hi Radek,
\nYes - Training offline is allowed, with the notebook providing inference. However please note that winners must include training and inference as a part of their winning submissions for review and validation.
", "rawMarkdown": "Hi Radek,\n\nYes - Training offline is allowed, with the notebook providing inference. However please note that winners must include training and inference as a part of their winning submissions for review and validation.", "votes": null }, { "id": "1795517", "postDate": "05/19/2022 21:06:08", "content": "Thank you very much for your reply, @addisonhoward, appreciate it! 🙂
", "rawMarkdown": "Thank you very much for your reply, @addisonhoward, appreciate it! 🙂", "votes": null }, { "id": "1795519", "postDate": "05/19/2022 21:08:31", "content": "Thank you very much, @yuanzhezhou!!!
", "rawMarkdown": "Thank you very much, @yuanzhezhou!!!", "votes": null }, { "id": "1795520", "postDate": "05/19/2022 21:08:49", "content": "Thank you @wcukierski!!!
", "rawMarkdown": "Thank you @wcukierski!!!", "votes": null }, { "id": "1801392", "postDate": "05/25/2022 17:19:10", "content": "Just to confirm, the 9 hour limit for GPUs / CPUs is ONLY for inference / predicting the test set? Our training scripts will be submitted as well, but do they need to be converted to .ipynb / something that can run on the kaggle platform? Or can it be something more traditional like repo with shell scripts included for reproducibility and those don't have a set time limit for executing the training script?
\nThis is my first kaggle competition, so I'm just not 100% what the regular practices are!
", "rawMarkdown": "Just to confirm, the 9 hour limit for GPUs / CPUs is ONLY for inference / predicting the test set? Our training scripts will be submitted as well, but do they need to be converted to .ipynb / something that can run on the kaggle platform? Or can it be something more traditional like repo with shell scripts included for reproducibility and those don't have a set time limit for executing the training script?\n\nThis is my first kaggle competition, so I'm just not 100% what the regular practices are!", "votes": null }, { "id": "1801434", "postDate": "05/25/2022 18:14:55", "content": "Welcome, @shypula! Indeed, the 9 hour limit only needs to include the inference on the test set. Your training scripts are only collected in the event you win a prize, or wish to voluntarily publish them.
", "rawMarkdown": "Welcome, @shypula! Indeed, the 9 hour limit only needs to include the inference on the test set. Your training scripts are only collected in the event you win a prize, or wish to voluntarily publish them.", "votes": null }, { "id": "1801643", "postDate": "05/26/2022 01:43:06", "content": "Sweet! And thanks!
", "rawMarkdown": "Sweet! And thanks!", "votes": null }, { "id": "1842724", "postDate": "07/04/2022 08:36:04", "content": "makes sense!
", "rawMarkdown": "makes sense!", "votes": null } ], "comments": [ { "id": 1795089, "author_name": "yuanzhezhou", "author_url": "", "post_date": "05/19/2022 11:59:43", "content": "yes, you can train your model else where. There will be about 20000 notebooks for test phase (check data tab).
", "votes": null, "replies": [ { "id": 1795519, "author_name": "radek1", "author_url": "", "post_date": "05/19/2022 21:08:31", "content": "Thank you very much, @yuanzhezhou!!!
", "votes": null, "replies": [] } ] }, { "id": 1795213, "author_name": "wcukierski", "author_url": "", "post_date": "05/19/2022 14:35:25", "content": "\n\ndo we know how much data will be made available to our model in the subsequent phases of the competition?
\n
The time limit remains the same for both stages.
", "votes": null, "replies": [ { "id": 1795520, "author_name": "radek1", "author_url": "", "post_date": "05/19/2022 21:08:49", "content": "Thank you @wcukierski!!!
", "votes": null, "replies": [] }, { "id": 1842724, "author_name": "abdulkadirguner", "author_url": "", "post_date": "07/04/2022 08:36:04", "content": "makes sense!
", "votes": null, "replies": [] } ] }, { "id": 1795287, "author_name": "addisonhoward", "author_url": "", "post_date": "05/19/2022 15:53:25", "content": "Hi Radek,
\nYes - Training offline is allowed, with the notebook providing inference. However please note that winners must include training and inference as a part of their winning submissions for review and validation.
", "votes": null, "replies": [ { "id": 1795517, "author_name": "radek1", "author_url": "", "post_date": "05/19/2022 21:06:08", "content": "Thank you very much for your reply, @addisonhoward, appreciate it! 🙂
", "votes": null, "replies": [] } ] }, { "id": 1801392, "author_name": "shypula", "author_url": "", "post_date": "05/25/2022 17:19:10", "content": "Just to confirm, the 9 hour limit for GPUs / CPUs is ONLY for inference / predicting the test set? Our training scripts will be submitted as well, but do they need to be converted to .ipynb / something that can run on the kaggle platform? Or can it be something more traditional like repo with shell scripts included for reproducibility and those don't have a set time limit for executing the training script?
\nThis is my first kaggle competition, so I'm just not 100% what the regular practices are!
", "votes": null, "replies": [ { "id": 1801434, "author_name": "wcukierski", "author_url": "", "post_date": "05/25/2022 18:14:55", "content": "Welcome, @shypula! Indeed, the 9 hour limit only needs to include the inference on the test set. Your training scripts are only collected in the event you win a prize, or wish to voluntarily publish them.
", "votes": null, "replies": [] }, { "id": 1801643, "author_name": "shypula", "author_url": "", "post_date": "05/26/2022 01:43:06", "content": "Sweet! And thanks!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1795075": "Hey!\n\nIf I understand correctly, in a code competition I can still train and experiment locally on my own machine. I can then upload trained models and perform inference via a kaggle kernel?\n\nMeaning, it's okay if I train the models elsewhere, not on Kaggle? My kernel can just do inference?\n\nBTW, the time limit is 9 hrs for the kernel. But how do we know how much data will be made available to our model in the subsequent phases of the competition?", "1795089": "yes, you can train your model else where. There will be about 20000 notebooks for test phase (check data tab).", "1795213": "> do we know how much data will be made available to our model in the subsequent phases of the competition?\n\nThe time limit remains the same for both stages.", "1795287": "Hi Radek,\n\nYes - Training offline is allowed, with the notebook providing inference. However please note that winners must include training and inference as a part of their winning submissions for review and validation.", "1795517": "Thank you very much for your reply, @addisonhoward, appreciate it! 🙂", "1795519": "Thank you very much, @yuanzhezhou!!!", "1795520": "Thank you @wcukierski!!!", "1801392": "Just to confirm, the 9 hour limit for GPUs / CPUs is ONLY for inference / predicting the test set? Our training scripts will be submitted as well, but do they need to be converted to .ipynb / something that can run on the kaggle platform? Or can it be something more traditional like repo with shell scripts included for reproducibility and those don't have a set time limit for executing the training script?\n\nThis is my first kaggle competition, so I'm just not 100% what the regular practices are!", "1801434": "Welcome, @shypula! Indeed, the 9 hour limit only needs to include the inference on the test set. Your training scripts are only collected in the event you win a prize, or wish to voluntarily publish them.", "1801643": "Sweet! And thanks!", "1842724": "makes sense!" }, "source": "meta" }