![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F991320%2F2ba55af2967c794cab220f0f88dfce70%2FScreen%20Shot%202019-12-27%20at%2012.24.29%20PM.png?generation=1577467570480087&alt=media\")
It seems to me that the parquet files takes too much time to read. For reference, reading test_image_data_0.parquet takes around 90 seconds while the actual inference is only 0.9 second. I feel like the 2hr limit on GPU kernel is a little bit unreasonable.
It seems to me that the parquet files takes too much time to read. For reference, reading test_image_data_0.parquet takes around 90 seconds while the actual inference is only 0.9 second. I feel like the 2hr limit on GPU kernel is a little bit unreasonable.
I agree, I had a few submissions which timed out too . Including the preprocessing and everything on the test dataset , 2 hours is a bit too less IMO.
", "rawMarkdown": "I agree, I had a few submissions which timed out too . Including the preprocessing and everything on the test dataset , 2 hours is a bit too less IMO.", "votes": null }, { "id": "704599", "postDate": "12/27/2019 17:19:19", "content": "Just curios where its written about 2 hours? I thought kernel limits are 9 hours (GPU)?
\n\nEdit: nvm found it \n
thanks for the info =)
", "rawMarkdown": "Just curios where its written about 2 hours? I thought kernel limits are 9 hours (GPU)? \n\nEdit: nvm found it \n\n\n\nthanks for the info =)", "votes": null }, { "id": "748276", "postDate": "02/17/2020 10:43:10", "content": "Anyone has an idea why loading the test parquet file with 3 rows takes 90 seconds? Does it take way more time for the full test or is there some overhead for few rows?
", "rawMarkdown": "Anyone has an idea why loading the test parquet file with 3 rows takes 90 seconds? Does it take way more time for the full test or is there some overhead for few rows?", "votes": null }, { "id": "748379", "postDate": "02/17/2020 13:14:24", "content": "for me it depends on the kernel (so it's random), sometimes reading the very small parquet file takes around 90 seconds but for some other kernels it takes less than 3 seconds.
\n\nIt's a bit painful because you need to wait about 6 min to be able to submit after a commit but I think you don't need to worry about how long it will take during submission, my inference time is less than 30 minutes even if I have to wait 6 min to commit on 12 examples. I guess it's not the same machine that perform the submission.
", "rawMarkdown": "for me it depends on the kernel (so it's random), sometimes reading the very small parquet file takes around 90 seconds but for some other kernels it takes less than 3 seconds.\n\nIt's a bit painful because you need to wait about 6 min to be able to submit after a commit but I think you don't need to worry about how long it will take during submission, my inference time is less than 30 minutes even if I have to wait 6 min to commit on 12 examples. I guess it's not the same machine that perform the submission.", "votes": null }, { "id": "748407", "postDate": "02/17/2020 13:52:25", "content": "Someone had mentioned that it's due to the large number of columns. The load time shouldn't scale linearly with the number of rows.
", "rawMarkdown": "Someone had mentioned that it's due to the large number of columns. The load time shouldn't scale linearly with the number of rows.", "votes": null }, { "id": "748461", "postDate": "02/17/2020 14:45:25", "content": "Thanks, it seems to be indeed the case.
", "rawMarkdown": "Thanks, it seems to be indeed the case.", "votes": null } ], "comments": [ { "id": 704195, "author_name": "p4rallax", "author_url": "", "post_date": "12/27/2019 06:38:55", "content": "I agree, I had a few submissions which timed out too . Including the preprocessing and everything on the test dataset , 2 hours is a bit too less IMO.
", "votes": null, "replies": [] }, { "id": 704599, "author_name": "drhabib", "author_url": "", "post_date": "12/27/2019 17:19:19", "content": "Just curios where its written about 2 hours? I thought kernel limits are 9 hours (GPU)?
\n\nEdit: nvm found it \n
thanks for the info =)
", "votes": null, "replies": [] }, { "id": 748276, "author_name": "philippsinger", "author_url": "", "post_date": "02/17/2020 10:43:10", "content": "Anyone has an idea why loading the test parquet file with 3 rows takes 90 seconds? Does it take way more time for the full test or is there some overhead for few rows?
", "votes": null, "replies": [ { "id": 748407, "author_name": "vaillant", "author_url": "", "post_date": "02/17/2020 13:52:25", "content": "Someone had mentioned that it's due to the large number of columns. The load time shouldn't scale linearly with the number of rows.
", "votes": null, "replies": [] }, { "id": 748461, "author_name": "philippsinger", "author_url": "", "post_date": "02/17/2020 14:45:25", "content": "Thanks, it seems to be indeed the case.
", "votes": null, "replies": [] } ] }, { "id": 748379, "author_name": "optimo", "author_url": "", "post_date": "02/17/2020 13:14:24", "content": "for me it depends on the kernel (so it's random), sometimes reading the very small parquet file takes around 90 seconds but for some other kernels it takes less than 3 seconds.
\n\nIt's a bit painful because you need to wait about 6 min to be able to submit after a commit but I think you don't need to worry about how long it will take during submission, my inference time is less than 30 minutes even if I have to wait 6 min to commit on 12 examples. I guess it's not the same machine that perform the submission.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "701822": "It seems to me that the parquet files takes too much time to read. For reference, reading `test_image_data_0.parquet` takes around 90 seconds while the actual inference is only 0.9 second. I feel like the 2hr limit on GPU kernel is a little bit unreasonable.", "704195": "I agree, I had a few submissions which timed out too . Including the preprocessing and everything on the test dataset , 2 hours is a bit too less IMO.", "704599": "Just curios where its written about 2 hours? I thought kernel limits are 9 hours (GPU)? \n\nEdit: nvm found it \n\n\n\nthanks for the info =)", "748276": "Anyone has an idea why loading the test parquet file with 3 rows takes 90 seconds? Does it take way more time for the full test or is there some overhead for few rows?", "748379": "for me it depends on the kernel (so it's random), sometimes reading the very small parquet file takes around 90 seconds but for some other kernels it takes less than 3 seconds.\n\nIt's a bit painful because you need to wait about 6 min to be able to submit after a commit but I think you don't need to worry about how long it will take during submission, my inference time is less than 30 minutes even if I have to wait 6 min to commit on 12 examples. I guess it's not the same machine that perform the submission.", "748407": "Someone had mentioned that it's due to the large number of columns. The load time shouldn't scale linearly with the number of rows.", "748461": "Thanks, it seems to be indeed the case." }, "source": "meta" }