![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2791153%2F2cc5494ce057fb1908497726e38457b0%2FScreen%20Shot%202019-12-20%20at%207.03.29%20AM.png?generation=1576843486774427&alt=media\")
I wanted to do quick implementation. Sadly, I have no GPU quota currently ;(
I also don't have GPU, so I'm using Google Colab, there's free Tesla P100, it's fast😏 . @hanjoonchoe \n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3876174%2F4c7ff06893e065365f99bbf55e3c71a8%2F2019-12-25_181812.png?generation=1577269117502686&alt=media\")
Yes, Colab is our savior!
", "rawMarkdown": "Yes, Colab is our savior!" }, { "id": 703913, "postDate": "2019-12-26T19:37:31.337Z", "content": "Colab have time limit. They will kick your process out for 90 minutes. If they offer GCP coupons, it would be better.
", "rawMarkdown": "Colab have time limit. They will kick your process out for 90 minutes. If they offer GCP coupons, it would be better.", "votes": 2 }, { "id": 703933, "postDate": "2019-12-26T20:42:11.453Z", "content": "I have known that it is allowed up to 12 hours maximum so far?
", "rawMarkdown": "I have known that it is allowed up to 12 hours maximum so far?" } ] }, { "id": 699409, "postDate": "2019-12-20T12:14:52.940Z", "content": "What model are you using? Maybe you can use a smaller model.
", "rawMarkdown": "What model are you using? Maybe you can use a smaller model. ", "votes": 1, "replies": [ { "id": 699410, "postDate": "2019-12-20T12:15:44.567Z", "content": "I built Resnet-18 customized. ;) Thank you for advice! 48 hours -> 78 hours something strange XD
", "rawMarkdown": "I built Resnet-18 customized. ;) Thank you for advice! 48 hours -> 78 hours something strange XD" }, { "id": 699590, "postDate": "2019-12-20T16:06:22.120Z", "content": "Resnet-18 is not a big model. Are you loading data from parquet? As per my testing, it is much much faster to load PNG images.
", "rawMarkdown": "Resnet-18 is not a big model. Are you loading data from parquet? As per my testing, it is much much faster to load PNG images.", "votes": 1 }, { "id": 699652, "postDate": "2019-12-20T18:00:38.100Z", "content": "Png is faster? I will try it :) (because it is redundant process once the values are loaded from the file. I think load directly from table is good way to try) \nI loaded image from feather it is much faster than parquet around 0.64 second compared to around 200 sec
\n\nI fix my coding style with the same feather format dataset. it tooks 24min per epoch with shallow cnn model now. ;)\nResnet-18 takes 2hours.
", "rawMarkdown": "Png is faster? I will try it :) (because it is redundant process once the values are loaded from the file. I think load directly from table is good way to try) \nI loaded image from feather it is much faster than parquet around 0.64 second compared to around 200 sec\n\nI fix my coding style with the same feather format dataset. it tooks 24min per epoch with shallow cnn model now. ;)\nResnet-18 takes 2hours." } ] }, { "id": 699405, "postDate": "2019-12-20T12:04:58.760Z", "content": "I wanted to do quick implementation. Sadly, I have no GPU quota currently ;(
Is this rerun competition(kernel competition) as I know? ;)
", "rawMarkdown": "Is this rerun competition(kernel competition) as I know? ;)" } ] } ], "comments": [ { "id": 702903, "author_name": "DiegoJohnson", "author_url": "", "post_date": "2019-12-25T10:19:01.397000", "content": "I also don't have GPU, so I'm using Google Colab, there's free Tesla P100, it's fast😏 . @hanjoonchoe \n
Yes, Colab is our savior!
", "votes": 0, "replies": [] }, { "id": 703913, "author_name": "Shangqiu Li", "author_url": "", "post_date": "2019-12-26T19:37:31.337000", "content": "Colab have time limit. They will kick your process out for 90 minutes. If they offer GCP coupons, it would be better.
", "votes": 2, "replies": [] }, { "id": 703933, "author_name": "Hanjoon Choe", "author_url": "", "post_date": "2019-12-26T20:42:11.453000", "content": "I have known that it is allowed up to 12 hours maximum so far?
", "votes": 0, "replies": [] } ] }, { "id": 699409, "author_name": "ibraheemmoosa", "author_url": "", "post_date": "2019-12-20T12:14:52.940000", "content": "What model are you using? Maybe you can use a smaller model.
", "votes": 1, "replies": [ { "id": 699410, "author_name": "Hanjoon Choe", "author_url": "", "post_date": "2019-12-20T12:15:44.567000", "content": "I built Resnet-18 customized. ;) Thank you for advice! 48 hours -> 78 hours something strange XD
", "votes": 0, "replies": [] }, { "id": 699590, "author_name": "ibraheemmoosa", "author_url": "", "post_date": "2019-12-20T16:06:22.120000", "content": "Resnet-18 is not a big model. Are you loading data from parquet? As per my testing, it is much much faster to load PNG images.
", "votes": 1, "replies": [] }, { "id": 699652, "author_name": "Hanjoon Choe", "author_url": "", "post_date": "2019-12-20T18:00:38.100000", "content": "Png is faster? I will try it :) (because it is redundant process once the values are loaded from the file. I think load directly from table is good way to try) \nI loaded image from feather it is much faster than parquet around 0.64 second compared to around 200 sec
\n\nI fix my coding style with the same feather format dataset. it tooks 24min per epoch with shallow cnn model now. ;)\nResnet-18 takes 2hours.
", "votes": 0, "replies": [] } ] }, { "id": 699437, "author_name": "", "author_url": "", "post_date": "2019-12-20T13:07:27.630000", "content": "", "votes": 1, "replies": [ { "id": 699440, "author_name": "Hanjoon Choe", "author_url": "", "post_date": "2019-12-20T13:09:09.747000", "content": "Is this rerun competition(kernel competition) as I know? ;)
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "702903": "I also don't have GPU, so I'm using Google Colab, there's free Tesla P100, it's fast😏 . @hanjoonchoe \n\n", "699409": "What model are you using? Maybe you can use a smaller model. ", "699405": "I wanted to do quick implementation. Sadly, I have no GPU quota currently ;(\n", "699437": "" } }