{ "id": 68711, "title": "keras starter kit (lb 0.892)", "url": "/competitions/quickdraw-doodle-recognition/discussion/68711", "author_name": "beluga", "post_date": "2018-10-16T11:22:39.404000", "votes": 35, "comment_count": 6, "views": 0, "content": "

Following Heng CherKeng's idea I will share keras benchmark kernels as I move along.

\n\n", "messages": [ { "id": 404775, "postDate": "2018-10-16T11:22:39.403Z", "content": "

Following Heng CherKeng's idea I will share keras benchmark kernels as I move along.

\n\n", "rawMarkdown": "Following Heng CherKeng's idea I will share keras benchmark kernels as I move along.\n\n* LB 0.002 [Silly benchmark for fun][1]\n* LB 0.771 [Black&White CNN][2]\n* LB 0.892 [Greyscale MobileNet][3]\n\n [1]: https://www.kaggle.com/gaborfodor/how-to-draw-an-owl-baseline-lb-0-002\n [2]: https://www.kaggle.com/gaborfodor/black-white-cnn-lb-0-77\n [3]: https://www.kaggle.com/gaborfodor/greyscale-mobilenet-lb-0-89", "votes": 34 }, { "id": 406157, "postDate": "2018-10-18T18:39:36.930Z", "content": "

@beluga - actually thank you for posting something like this even with such a high score. When it comes to competitions like this - I think there is a lot to learn. I did not realize that models like MobileNet could be used out of the box and also I did not expect to see such variation between epochs for validation during training. Up to this point whenever I have seen that I stopped training thinking something was wrong. Appreciate the \"knowledge transfer\" :)

", "rawMarkdown": "@beluga - actually thank you for posting something like this even with such a high score. When it comes to competitions like this - I think there is a lot to learn. I did not realize that models like MobileNet could be used out of the box and also I did not expect to see such variation between epochs for validation during training. Up to this point whenever I have seen that I stopped training thinking something was wrong. Appreciate the \"knowledge transfer\" :)", "votes": 3 }, { "id": 406144, "postDate": "2018-10-18T18:18:10.297Z", "content": "

Ok the last one got better than I was expected. I will slow down a bit :)

", "rawMarkdown": "Ok the last one got better than I was expected. I will slow down a bit :)", "votes": 1 }, { "id": 406164, "postDate": "2018-10-18T18:49:10.910Z", "content": "

wow, 0892 starting kit? Thanks @beluga, you're the guy!

", "rawMarkdown": "wow, 0892 starting kit? Thanks @beluga, you're the guy!", "votes": 2 }, { "id": 406312, "postDate": "2018-10-19T03:08:46.490Z", "content": "

I'd been exceeding the time limit on an Kaggle kernel trying the MobileNet, and you come like a saviour. Thanks for that.

", "rawMarkdown": "I'd been exceeding the time limit on an Kaggle kernel trying the MobileNet, and you come like a saviour. Thanks for that." }, { "id": 405579, "postDate": "2018-10-17T18:36:00.230Z", "content": "

Good luck.

", "rawMarkdown": "Good luck." }, { "id": 406408, "postDate": "2018-10-19T07:42:02.453Z", "content": "

@beluga, thanks very much.

", "rawMarkdown": " @beluga, thanks very much." } ], "comments": [ { "id": 406157, "author_name": "RDizzl3", "author_url": "", "post_date": "2018-10-18T18:39:36.930000", "content": "

@beluga - actually thank you for posting something like this even with such a high score. When it comes to competitions like this - I think there is a lot to learn. I did not realize that models like MobileNet could be used out of the box and also I did not expect to see such variation between epochs for validation during training. Up to this point whenever I have seen that I stopped training thinking something was wrong. Appreciate the \"knowledge transfer\" :)

", "votes": 3, "replies": [] }, { "id": 406144, "author_name": "beluga", "author_url": "", "post_date": "2018-10-18T18:18:10.297000", "content": "

Ok the last one got better than I was expected. I will slow down a bit :)

", "votes": 1, "replies": [] }, { "id": 406164, "author_name": "Giba", "author_url": "", "post_date": "2018-10-18T18:49:10.910000", "content": "

wow, 0892 starting kit? Thanks @beluga, you're the guy!

", "votes": 2, "replies": [] }, { "id": 406312, "author_name": "Rajath", "author_url": "", "post_date": "2018-10-19T03:08:46.490000", "content": "

I'd been exceeding the time limit on an Kaggle kernel trying the MobileNet, and you come like a saviour. Thanks for that.

", "votes": 0, "replies": [] }, { "id": 405579, "author_name": "Tomas Rasymas", "author_url": "", "post_date": "2018-10-17T18:36:00.230000", "content": "

Good luck.

", "votes": 0, "replies": [] }, { "id": 406408, "author_name": "yyqing", "author_url": "", "post_date": "2018-10-19T07:42:02.453000", "content": "

@beluga, thanks very much.

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "404775": "Following Heng CherKeng's idea I will share keras benchmark kernels as I move along.\n\n* LB 0.002 [Silly benchmark for fun][1]\n* LB 0.771 [Black&White CNN][2]\n* LB 0.892 [Greyscale MobileNet][3]\n\n [1]: https://www.kaggle.com/gaborfodor/how-to-draw-an-owl-baseline-lb-0-002\n [2]: https://www.kaggle.com/gaborfodor/black-white-cnn-lb-0-77\n [3]: https://www.kaggle.com/gaborfodor/greyscale-mobilenet-lb-0-89", "406157": "@beluga - actually thank you for posting something like this even with such a high score. When it comes to competitions like this - I think there is a lot to learn. I did not realize that models like MobileNet could be used out of the box and also I did not expect to see such variation between epochs for validation during training. Up to this point whenever I have seen that I stopped training thinking something was wrong. Appreciate the \"knowledge transfer\" :)", "406144": "Ok the last one got better than I was expected. I will slow down a bit :)", "406164": "wow, 0892 starting kit? Thanks @beluga, you're the guy!", "406312": "I'd been exceeding the time limit on an Kaggle kernel trying the MobileNet, and you come like a saviour. Thanks for that.", "405579": "Good luck.", "406408": " @beluga, thanks very much." } }