{ "id": 391674, "title": "Code Competition clarification", "url": "/competitions/asl-signs/discussion/391674", "author_name": "Maxim B", "post_date": "2023-03-02T08:46:14.485000", "votes": 3, "comment_count": 7, "views": 0, "content": "
\n

This is a Code Competition
\n Submissions to this competition must be made through Notebooks. In order for the \"Submit\" button to be active after a commit, the following conditions must be met:

\n \n
\n

Does this mean that our model also needs to train within a 9-hour notebook? Or can we train offline (and have a minimal notebook that just loads our tflite model)?

\n

or in otherwords is a notebook with the following single line of code allowed?
\n!zip submission.zip /kaggle/input/private-dataset/model.tflite

", "messages": [ { "id": 2165504, "postDate": "2023-03-02T08:46:14.487Z", "content": "
\n

This is a Code Competition
\n Submissions to this competition must be made through Notebooks. In order for the \"Submit\" button to be active after a commit, the following conditions must be met:

\n \n
\n

Does this mean that our model also needs to train within a 9-hour notebook? Or can we train offline (and have a minimal notebook that just loads our tflite model)?

\n

or in otherwords is a notebook with the following single line of code allowed?
\n!zip submission.zip /kaggle/input/private-dataset/model.tflite

", "rawMarkdown": ">This is a Code Competition\nSubmissions to this competition must be made through Notebooks. In order for the \"Submit\" button to be active after a commit, the following conditions must be met:\n* CPU Notebook <= 9 hours run-time\n* GPU Notebook <= 9 hours run-time\n* Internet access disabled\n* Freely & publicly available external data is allowed, including pre-trained models\n* Submission file must be named submission.zip.\nPlease see the Code Competition FAQ for more information on how to submit. And review the code debugging doc if you are encountering submission errors.\n\nDoes this mean that our model also needs to **train** within a 9-hour notebook? Or can we train offline (and have a minimal notebook that just loads our tflite model)?\n\n**or in otherwords is a notebook with the following single line of code allowed?**\n`!zip submission.zip /kaggle/input/private-dataset/model.tflite`", "votes": 3 }, { "id": 2206963, "postDate": "2023-04-03T03:19:36.917Z", "content": "

I would like to ask on which GPU (P100?) the model ran when they apply inference externally. Should I consider the lack of memory?

", "rawMarkdown": "I would like to ask on which GPU (P100?) the model ran when they apply inference externally. Should I consider the lack of memory?", "votes": 1 }, { "id": 2166409, "postDate": "2023-03-02T19:19:02.820Z", "content": "

The Code Competition rules are boilerplate. I'd say they don't really apply to this competition, which really is a Model Competition. I hope Kaggle team creates a new template for Model Competitions with its own FAQ.

\n

A one-liner notebook that zips a model file would be fine. (Though I haven't tested it). And I believe you can take any notebook or dataset that has ALREADY run, with a 'submission.zip' file and click 'submit' and it will work if the zip contains a working tflite model.

\n

The real rules are on the Evaluation page:

\n
\n

Your model must also require less than 40 MB in memory and perform inference with less than 100 milliseconds of latency per video. Expect to see approximately 40,000 videos in the test set. We allow an additional 10 minute buffer for loading the data and miscellaneous overhead.

\n
\n

So the notebook time to generate the submission.zip file shouldn't matter. The inference needs to finish within about 75 minutes. Give or take (~40,000*0.1 seconds plus 600 seconds buffer).

", "rawMarkdown": "The Code Competition rules are boilerplate. I'd say they don't really apply to this competition, which really is a Model Competition. I hope Kaggle team creates a new template for Model Competitions with its own FAQ.\n\nA one-liner notebook that zips a model file would be fine. (Though I haven't tested it). And I believe you can take any notebook or dataset that has ALREADY run, with a 'submission.zip' file and click 'submit' and it will work if the zip contains a working tflite model.\n\nThe **real** rules are on [the Evaluation page](https://www.kaggle.com/competitions/asl-signs/overview/evaluation):\n>Your model must also require less than 40 MB in memory and perform inference with less than 100 milliseconds of latency per video. Expect to see approximately 40,000 videos in the test set. We allow an additional 10 minute buffer for loading the data and miscellaneous overhead.\n\nSo the notebook time to generate the submission.zip file shouldn't matter. The inference needs to finish within about 75 minutes. Give or take (~40,000*0.1 seconds plus 600 seconds buffer).", "votes": 1, "replies": [ { "id": 2170616, "postDate": "2023-03-06T06:33:00.960Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 2197783, "postDate": "2023-03-26T12:22:08.807Z", "content": "

@sohier please help to clarify. Thanks

", "rawMarkdown": "@sohier please help to clarify. Thanks" } ] } ] }, { "id": 2165779, "postDate": "2023-03-02T12:59:14.113Z", "content": "

@mxbonn you can train offline. inference time is with in 9 hours

", "rawMarkdown": "@mxbonn you can train offline. inference time is with in 9 hours", "replies": [ { "id": 2165882, "postDate": "2023-03-02T14:17:08.970Z", "content": "

But the actual inference does not happen in the notebook, you submit a tflite model on which they apply inference externally.

\n

So do you think a notebook with this one-liner is allowed?:
\n!zip submission.zip /kaggle/input/private-dataset/model.tflite

", "rawMarkdown": "But the actual inference does not happen in the notebook, you submit a tflite model on which they apply inference externally.\n\nSo do you think a notebook with this one-liner is allowed?:\n`!zip submission.zip /kaggle/input/private-dataset/model.tflite`", "replies": [ { "id": 2166187, "postDate": "2023-03-02T17:08:59.577Z", "content": "

That's what I've been using so far. (Although my training is comfortably under 9 hours, so I could easily do the whole thing in a single notebook if required.)

\n

In the unlikely event that I finish anywhere near the top, I'll make my complete code available.

", "rawMarkdown": "That's what I've been using so far. (Although my training is comfortably under 9 hours, so I could easily do the whole thing in a single notebook if required.)\n\nIn the unlikely event that I finish anywhere near the top, I'll make my complete code available.", "votes": 1 } ] } ] } ], "comments": [ { "id": 2206963, "author_name": "Jackson You", "author_url": "", "post_date": "2023-04-03T03:19:36.917000", "content": "

I would like to ask on which GPU (P100?) the model ran when they apply inference externally. Should I consider the lack of memory?

", "votes": 1, "replies": [] }, { "id": 2166409, "author_name": "Robert Hatch", "author_url": "", "post_date": "2023-03-02T19:19:02.820000", "content": "

The Code Competition rules are boilerplate. I'd say they don't really apply to this competition, which really is a Model Competition. I hope Kaggle team creates a new template for Model Competitions with its own FAQ.

\n

A one-liner notebook that zips a model file would be fine. (Though I haven't tested it). And I believe you can take any notebook or dataset that has ALREADY run, with a 'submission.zip' file and click 'submit' and it will work if the zip contains a working tflite model.

\n

The real rules are on the Evaluation page:

\n
\n

Your model must also require less than 40 MB in memory and perform inference with less than 100 milliseconds of latency per video. Expect to see approximately 40,000 videos in the test set. We allow an additional 10 minute buffer for loading the data and miscellaneous overhead.

\n
\n

So the notebook time to generate the submission.zip file shouldn't matter. The inference needs to finish within about 75 minutes. Give or take (~40,000*0.1 seconds plus 600 seconds buffer).

", "votes": 1, "replies": [ { "id": 2170616, "author_name": "", "author_url": "", "post_date": "2023-03-06T06:33:00.960000", "content": "", "votes": 0, "replies": [ { "id": 2197783, "author_name": "Just A game on your lips", "author_url": "", "post_date": "2023-03-26T12:22:08.807000", "content": "

@sohier please help to clarify. Thanks

", "votes": 0, "replies": [] } ] } ] }, { "id": 2165779, "author_name": "SeshuRaju 🧘‍♂️", "author_url": "", "post_date": "2023-03-02T12:59:14.113000", "content": "

@mxbonn you can train offline. inference time is with in 9 hours

", "votes": 0, "replies": [ { "id": 2165882, "author_name": "Maxim B", "author_url": "", "post_date": "2023-03-02T14:17:08.970000", "content": "

But the actual inference does not happen in the notebook, you submit a tflite model on which they apply inference externally.

\n

So do you think a notebook with this one-liner is allowed?:
\n!zip submission.zip /kaggle/input/private-dataset/model.tflite

", "votes": 0, "replies": [ { "id": 2166187, "author_name": "Andrew", "author_url": "", "post_date": "2023-03-02T17:08:59.577000", "content": "

That's what I've been using so far. (Although my training is comfortably under 9 hours, so I could easily do the whole thing in a single notebook if required.)

\n

In the unlikely event that I finish anywhere near the top, I'll make my complete code available.

", "votes": 1, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2165504": ">This is a Code Competition\nSubmissions to this competition must be made through Notebooks. In order for the \"Submit\" button to be active after a commit, the following conditions must be met:\n* CPU Notebook <= 9 hours run-time\n* GPU Notebook <= 9 hours run-time\n* Internet access disabled\n* Freely & publicly available external data is allowed, including pre-trained models\n* Submission file must be named submission.zip.\nPlease see the Code Competition FAQ for more information on how to submit. And review the code debugging doc if you are encountering submission errors.\n\nDoes this mean that our model also needs to **train** within a 9-hour notebook? Or can we train offline (and have a minimal notebook that just loads our tflite model)?\n\n**or in otherwords is a notebook with the following single line of code allowed?**\n`!zip submission.zip /kaggle/input/private-dataset/model.tflite`", "2206963": "I would like to ask on which GPU (P100?) the model ran when they apply inference externally. Should I consider the lack of memory?", "2166409": "The Code Competition rules are boilerplate. I'd say they don't really apply to this competition, which really is a Model Competition. I hope Kaggle team creates a new template for Model Competitions with its own FAQ.\n\nA one-liner notebook that zips a model file would be fine. (Though I haven't tested it). And I believe you can take any notebook or dataset that has ALREADY run, with a 'submission.zip' file and click 'submit' and it will work if the zip contains a working tflite model.\n\nThe **real** rules are on [the Evaluation page](https://www.kaggle.com/competitions/asl-signs/overview/evaluation):\n>Your model must also require less than 40 MB in memory and perform inference with less than 100 milliseconds of latency per video. Expect to see approximately 40,000 videos in the test set. We allow an additional 10 minute buffer for loading the data and miscellaneous overhead.\n\nSo the notebook time to generate the submission.zip file shouldn't matter. The inference needs to finish within about 75 minutes. Give or take (~40,000*0.1 seconds plus 600 seconds buffer).", "2165779": "@mxbonn you can train offline. inference time is with in 9 hours" } }