I'm using a CPU-only kernel that runs in ~270 seconds on commit for the 400 public test samples, but still generates a timeout error when run on the test set. I haven't been able to identify the issue. If I force the kernel to process only 1000 videos, I get a different error (I assume from having an incorrectly sized submission file) after about 20 minutes.
\n\nIs anyone else having the same issue?\nCould it be related to the kernel using only the CPU (GPU option turned off)?
\n\nEdit: I'm not using Neural Networks, so other than some OpenCv operations, I doubt I'd see any performance improvements from using a GPU.
", "messages": [ { "id": 727425, "postDate": "2020-01-23T18:01:09.140Z", "content": "So far I have only used the CPU. My timing for the 400 is around the same as yours running the script step by step. When I commit its often couple of hours before I get the completed message - but upon opening the version it shows that it was completed in around the 300 seconds I expected. This issue only occurs when I use ffmpeg to generate an audio file. Most of my submissions attempts have been failures - but timeout has never been the issue. I added a timing print for the ffmpeg loop and on a random basis some of the loops took 30+ seconds when they should have taken only a fraction - something I was doing appeared to be generating a timeout situation (I think).
\n\nSo my experience would suggest that cpu is not your issue.
", "rawMarkdown": "So far I have only used the CPU. My timing for the 400 is around the same as yours running the script step by step. When I commit its often couple of hours before I get the completed message - but upon opening the version it shows that it was completed in around the 300 seconds I expected. This issue only occurs when I use ffmpeg to generate an audio file. Most of my submissions attempts have been failures - but timeout has never been the issue. I added a timing print for the ffmpeg loop and on a random basis some of the loops took 30+ seconds when they should have taken only a fraction - something I was doing appeared to be generating a timeout situation (I think).\n\nSo my experience would suggest that cpu is not your issue.", "votes": 1 }, { "id": 725706, "postDate": "2020-01-22T12:16:47.823Z", "content": "I'm using a CPU-only kernel that runs in ~270 seconds on commit for the 400 public test samples, but still generates a timeout error when run on the test set. I haven't been able to identify the issue. If I force the kernel to process only 1000 videos, I get a different error (I assume from having an incorrectly sized submission file) after about 20 minutes.
\n\nIs anyone else having the same issue?\nCould it be related to the kernel using only the CPU (GPU option turned off)?
\n\nEdit: I'm not using Neural Networks, so other than some OpenCv operations, I doubt I'd see any performance improvements from using a GPU.
", "rawMarkdown": "I'm using a CPU-only kernel that runs in ~270 seconds on commit for the 400 public test samples, but still generates a timeout error when run on the test set. I haven't been able to identify the issue. If I force the kernel to process only 1000 videos, I get a different error (I assume from having an incorrectly sized submission file) after about 20 minutes.\n\nIs anyone else having the same issue?\nCould it be related to the kernel using only the CPU (GPU option turned off)?\n\nEdit: I'm not using Neural Networks, so other than some OpenCv operations, I doubt I'd see any performance improvements from using a GPU." }, { "id": 725741, "postDate": "2020-01-22T13:09:44.123Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 725816, "postDate": "2020-01-22T14:13:24.703Z", "content": "No neural network, for now just a pre-trained logistic regression classifier with some features I'm extracting (4 features per video and 1 second of processing per video). I'm not expecting some brilliant performance, but I find it odd that it doesn't work.
", "rawMarkdown": "No neural network, for now just a pre-trained logistic regression classifier with some features I'm extracting (4 features per video and 1 second of processing per video). I'm not expecting some brilliant performance, but I find it odd that it doesn't work. " } ] } ], "comments": [ { "id": 727425, "author_name": "PC Jimmmy", "author_url": "", "post_date": "2020-01-23T18:01:09.140000", "content": "So far I have only used the CPU. My timing for the 400 is around the same as yours running the script step by step. When I commit its often couple of hours before I get the completed message - but upon opening the version it shows that it was completed in around the 300 seconds I expected. This issue only occurs when I use ffmpeg to generate an audio file. Most of my submissions attempts have been failures - but timeout has never been the issue. I added a timing print for the ffmpeg loop and on a random basis some of the loops took 30+ seconds when they should have taken only a fraction - something I was doing appeared to be generating a timeout situation (I think).
\n\nSo my experience would suggest that cpu is not your issue.
", "votes": 1, "replies": [] }, { "id": 725741, "author_name": "", "author_url": "", "post_date": "2020-01-22T13:09:44.123000", "content": "", "votes": 0, "replies": [ { "id": 725816, "author_name": "Camilo Martinez", "author_url": "", "post_date": "2020-01-22T14:13:24.703000", "content": "No neural network, for now just a pre-trained logistic regression classifier with some features I'm extracting (4 features per video and 1 second of processing per video). I'm not expecting some brilliant performance, but I find it odd that it doesn't work.
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "727425": "So far I have only used the CPU. My timing for the 400 is around the same as yours running the script step by step. When I commit its often couple of hours before I get the completed message - but upon opening the version it shows that it was completed in around the 300 seconds I expected. This issue only occurs when I use ffmpeg to generate an audio file. Most of my submissions attempts have been failures - but timeout has never been the issue. I added a timing print for the ffmpeg loop and on a random basis some of the loops took 30+ seconds when they should have taken only a fraction - something I was doing appeared to be generating a timeout situation (I think).\n\nSo my experience would suggest that cpu is not your issue.", "725706": "I'm using a CPU-only kernel that runs in ~270 seconds on commit for the 400 public test samples, but still generates a timeout error when run on the test set. I haven't been able to identify the issue. If I force the kernel to process only 1000 videos, I get a different error (I assume from having an incorrectly sized submission file) after about 20 minutes.\n\nIs anyone else having the same issue?\nCould it be related to the kernel using only the CPU (GPU option turned off)?\n\nEdit: I'm not using Neural Networks, so other than some OpenCv operations, I doubt I'd see any performance improvements from using a GPU.", "725741": "" } }