Here is what happened:
\nDoes anyone know how to handle this? What sort of things can I do to reduce RAM or disk utilization?
", "messages": [ { "id": "1391662", "postDate": "07/17/2021 21:37:15", "content": "Here is what happened:
\nDoes anyone know how to handle this? What sort of things can I do to reduce RAM or disk utilization?
", "rawMarkdown": "Here is what happened:\n1. My notebook worked before. I had one model for classification and one for detection, each with a 70/30 train/valid split\n2. I upgraded to using 5-fold cross-validation (I use weighted box fusion to combine the detection predictions. Might this be a problem? Does it take a lot of ram?)\n3. It works when I run the notebook locally, but when I submit I get this error \"notebook exceeded allowed compute\"\n\nDoes anyone know how to handle this? What sort of things can I do to reduce RAM or disk utilization?", "votes": null }, { "id": "1395789", "postDate": "07/21/2021 14:34:51", "content": "there is run time limit.
", "rawMarkdown": "there is run time limit.", "votes": null }, { "id": "1397070", "postDate": "07/22/2021 18:24:01", "content": "What is the time limit?
", "rawMarkdown": "What is the time limit?", "votes": null }, { "id": "1397580", "postDate": "07/23/2021 10:08:38", "content": "Time limit only matters if you had to wait some hours before the submission error appeared. Time limit is given either by the competition rules or general resource limitations from the administration. For this competition the rules say:
\n\n\nCPU Notebook <= 9 hours run-time
\n
\n GPU Notebook <= 9 hours run-time
I don't know how your code looks like, but usually your submitted code is run on a hidden data set. If you get the error notice after just a couple of minutes, it is likely that your method requires too much RAM. You can only avoid this by trial and error, try some of these general approaches:
\n-delete unused variables
\n-check for loops that have exponential effect
\n-look for possible batch processing
\n-avoid inefficient techniques in general
\n-it's also possible that you have a constructional fault and your locally run code only worked by chance
As you said it worked before (you mean also submitting ?), it is quite likely that you have to adjust the cross-validation process. I don't know which framework you use, but e.g. if it is sklearn.cross_validate() you may adjust the number of parallel jobs ...cross_validate(..., pre_dispatchint=n_jobs ,..).
@vdefont, hope this will help, otherwise feel free to specify the code you used.
", "rawMarkdown": "Time limit only matters if you had to wait some hours before the submission error appeared. Time limit is given either by the competition rules or general resource limitations from the administration. For this competition the rules say:\n> CPU Notebook <= 9 hours run-time\nGPU Notebook <= 9 hours run-time\n\nI don't know how your code looks like, but usually your submitted code is run on a hidden data set. If you get the error notice after just a couple of minutes, it is likely that your method requires too much RAM. You can only avoid this by trial and error, try some of these general approaches:\n\n-delete unused variables\n-check for loops that have exponential effect\n-look for possible batch processing\n-avoid inefficient techniques in general\n-it's also possible that you have a constructional fault and your locally run code only worked by chance \n\nAs you said it worked before (you mean also submitting ?), it is quite likely that you have to adjust the cross-validation process. I don't know which framework you use, but e.g. if it is `sklearn.cross_validate()` you may adjust the number of parallel jobs `...cross_validate(..., pre_dispatchint=n_jobs ,..)`.\n\n@vdefont, hope this will help, otherwise feel free to specify the code you used.", "votes": null } ], "comments": [ { "id": 1395789, "author_name": "dragonzhang", "author_url": "", "post_date": "07/21/2021 14:34:51", "content": "there is run time limit.
", "votes": null, "replies": [ { "id": 1397070, "author_name": "vdefont", "author_url": "", "post_date": "07/22/2021 18:24:01", "content": "What is the time limit?
", "votes": null, "replies": [] } ] }, { "id": 1397580, "author_name": "alexanderbader", "author_url": "", "post_date": "07/23/2021 10:08:38", "content": "Time limit only matters if you had to wait some hours before the submission error appeared. Time limit is given either by the competition rules or general resource limitations from the administration. For this competition the rules say:
\n\n\nCPU Notebook <= 9 hours run-time
\n
\n GPU Notebook <= 9 hours run-time
I don't know how your code looks like, but usually your submitted code is run on a hidden data set. If you get the error notice after just a couple of minutes, it is likely that your method requires too much RAM. You can only avoid this by trial and error, try some of these general approaches:
\n-delete unused variables
\n-check for loops that have exponential effect
\n-look for possible batch processing
\n-avoid inefficient techniques in general
\n-it's also possible that you have a constructional fault and your locally run code only worked by chance
As you said it worked before (you mean also submitting ?), it is quite likely that you have to adjust the cross-validation process. I don't know which framework you use, but e.g. if it is sklearn.cross_validate() you may adjust the number of parallel jobs ...cross_validate(..., pre_dispatchint=n_jobs ,..).
@vdefont, hope this will help, otherwise feel free to specify the code you used.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1391662": "Here is what happened:\n1. My notebook worked before. I had one model for classification and one for detection, each with a 70/30 train/valid split\n2. I upgraded to using 5-fold cross-validation (I use weighted box fusion to combine the detection predictions. Might this be a problem? Does it take a lot of ram?)\n3. It works when I run the notebook locally, but when I submit I get this error \"notebook exceeded allowed compute\"\n\nDoes anyone know how to handle this? What sort of things can I do to reduce RAM or disk utilization?", "1395789": "there is run time limit.", "1397070": "What is the time limit?", "1397580": "Time limit only matters if you had to wait some hours before the submission error appeared. Time limit is given either by the competition rules or general resource limitations from the administration. For this competition the rules say:\n> CPU Notebook <= 9 hours run-time\nGPU Notebook <= 9 hours run-time\n\nI don't know how your code looks like, but usually your submitted code is run on a hidden data set. If you get the error notice after just a couple of minutes, it is likely that your method requires too much RAM. You can only avoid this by trial and error, try some of these general approaches:\n\n-delete unused variables\n-check for loops that have exponential effect\n-look for possible batch processing\n-avoid inefficient techniques in general\n-it's also possible that you have a constructional fault and your locally run code only worked by chance \n\nAs you said it worked before (you mean also submitting ?), it is quite likely that you have to adjust the cross-validation process. I don't know which framework you use, but e.g. if it is `sklearn.cross_validate()` you may adjust the number of parallel jobs `...cross_validate(..., pre_dispatchint=n_jobs ,..)`.\n\n@vdefont, hope this will help, otherwise feel free to specify the code you used." }, "source": "meta" }