{ "id": 400208, "title": "Submission throws a Timeout excpetion", "url": "/competitions/birdclef-2023/discussion/400208", "author_name": "", "post_date": "2023-04-07T08:26:09.822923700Z", "votes": null, "comment_count": 1, "views": 0, "content": "

I'm using a built-in infrastructure for audio training and inference. The infrastructure license allows commercial uses so I am allowed to use it.
\nHowever, when I try to submit my results, the kernel throws a Timeout exception. I tried to check the matter and found that feeding a 5-second sample into my pre-trained model and outputting the predictions takes 2 seconds. This is problematic because one file has 120 samples and there are 200 files in the test, so it will take 2X120X200 = 48000 seconds to complete the test which is almost 13 hours!
\nDoes anyone have any suggestions on what I can do about it? I am very frustrated… I also try using multithreading processes but it reduced the time by half (reduced time from 13 hours to ~6 hours).
\nUsing this infrastructure is critical for me since it is comfortable to use and mainly because it contains state-of-the-art audio networks which have the potential to produce great results.
\nIn addition, I wanted to know why it is forbidden to use the GPU during the submission and why the limitation for kernel running is =<120 minutes. If it was possible I wouldn't have this problem.

\n

Thanks

", "messages": [ { "id": "2213000", "postDate": "04/07/2023 08:26:09", "content": "

I'm using a built-in infrastructure for audio training and inference. The infrastructure license allows commercial uses so I am allowed to use it.
\nHowever, when I try to submit my results, the kernel throws a Timeout exception. I tried to check the matter and found that feeding a 5-second sample into my pre-trained model and outputting the predictions takes 2 seconds. This is problematic because one file has 120 samples and there are 200 files in the test, so it will take 2X120X200 = 48000 seconds to complete the test which is almost 13 hours!
\nDoes anyone have any suggestions on what I can do about it? I am very frustrated… I also try using multithreading processes but it reduced the time by half (reduced time from 13 hours to ~6 hours).
\nUsing this infrastructure is critical for me since it is comfortable to use and mainly because it contains state-of-the-art audio networks which have the potential to produce great results.
\nIn addition, I wanted to know why it is forbidden to use the GPU during the submission and why the limitation for kernel running is =<120 minutes. If it was possible I wouldn't have this problem.

\n

Thanks

", "rawMarkdown": "I'm using a built-in infrastructure for audio training and inference. The infrastructure license allows commercial uses so I am allowed to use it.\nHowever, when I try to submit my results, the kernel throws a Timeout exception. I tried to check the matter and found that feeding a 5-second sample into my pre-trained model and outputting the predictions takes 2 seconds. This is problematic because one file has 120 samples and there are 200 files in the test, so it will take 2X120X200 = 48000 seconds to complete the test which is almost 13 hours!\nDoes anyone have any suggestions on what I can do about it? I am very frustrated... I also try using multithreading processes but it reduced the time by half (reduced time from 13 hours to ~6 hours).\nUsing this infrastructure is critical for me since it is comfortable to use and mainly because it contains state-of-the-art audio networks which have the potential to produce great results.\nIn addition, I wanted to know why it is forbidden to use the GPU during the submission and why the limitation for kernel running is =<120 minutes. If it was possible I wouldn't have this problem. \n\nThanks", "votes": null }, { "id": "2216046", "postDate": "04/09/2023 18:21:58", "content": "

Hi!
\nI also want to know the answer to your first question, but for your second question, you can take a look at this discussion.

", "rawMarkdown": "Hi!\nI also want to know the answer to your first question, but for your second question, you can take a look at this [discussion](https://www.kaggle.com/competitions/birdclef-2023/discussion/393059).", "votes": null } ], "comments": [ { "id": 2216046, "author_name": "aryankhatana", "author_url": "", "post_date": "04/09/2023 18:21:58", "content": "

Hi!
\nI also want to know the answer to your first question, but for your second question, you can take a look at this discussion.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2213000": "I'm using a built-in infrastructure for audio training and inference. The infrastructure license allows commercial uses so I am allowed to use it.\nHowever, when I try to submit my results, the kernel throws a Timeout exception. I tried to check the matter and found that feeding a 5-second sample into my pre-trained model and outputting the predictions takes 2 seconds. This is problematic because one file has 120 samples and there are 200 files in the test, so it will take 2X120X200 = 48000 seconds to complete the test which is almost 13 hours!\nDoes anyone have any suggestions on what I can do about it? I am very frustrated... I also try using multithreading processes but it reduced the time by half (reduced time from 13 hours to ~6 hours).\nUsing this infrastructure is critical for me since it is comfortable to use and mainly because it contains state-of-the-art audio networks which have the potential to produce great results.\nIn addition, I wanted to know why it is forbidden to use the GPU during the submission and why the limitation for kernel running is =<120 minutes. If it was possible I wouldn't have this problem. \n\nThanks", "2216046": "Hi!\nI also want to know the answer to your first question, but for your second question, you can take a look at this [discussion](https://www.kaggle.com/competitions/birdclef-2023/discussion/393059)." }, "source": "meta" }