I've tried many, many times, and by any ways I can't reach the score of 0.99 or above on the LB. I would like to ask those teachers about 0.99 points, how does this work? thanks.🙏 🙏
", "messages": [ { "id": "695465", "postDate": "12/15/2019 07:28:31", "content": "I've tried many, many times, and by any ways I can't reach the score of 0.99 or above on the LB. I would like to ask those teachers about 0.99 points, how does this work? thanks.🙏 🙏
", "rawMarkdown": "I've tried many, many times, and by any ways I can't reach the score of 0.99 or above on the LB. I would like to ask those teachers about 0.99 points, how does this work? thanks.🙏 🙏", "votes": null }, { "id": "695466", "postDate": "12/15/2019 07:30:03", "content": "Now my highest score on LB is 0.98940....😫
", "rawMarkdown": "Now my highest score on LB is 0.98940....😫", "votes": null }, { "id": "695480", "postDate": "12/15/2019 08:05:04", "content": "I am sure that you must have tried all those fancy techniques( Earlystopping, reducing learning rate, ImageDataGenerator etc.) since your score is 0.989.\n I have experienced some kind of a weird thing happening here on kaggle, as I ran same code in two different times and got 0.988 in one and 0.992 in the other !\nSo I would suggest that you run your code again and see for these changing public scores for yourself.
", "rawMarkdown": "I am sure that you must have tried all those fancy techniques( Earlystopping, reducing learning rate, ImageDataGenerator etc.) since your score is 0.989.\n I have experienced some kind of a weird thing happening here on kaggle, as I ran same code in two different times and got 0.988 in one and 0.992 in the other !\nSo I would suggest that you run your code again and see for these changing public scores for yourself.", "votes": null }, { "id": "695482", "postDate": "12/15/2019 08:09:42", "content": "Did your score change this time?
", "rawMarkdown": "Did your score change this time?", "votes": null }, { "id": "695504", "postDate": "12/15/2019 08:58:45", "content": "I also found that, the scores are different from different codes, I will try again, thank you!👍
", "rawMarkdown": "I also found that, the scores are different from different codes, I will try again, thank you!👍", "votes": null }, { "id": "695505", "postDate": "12/15/2019 09:01:42", "content": "nothing happened.😭
", "rawMarkdown": "nothing happened.😭", "votes": null }, { "id": "695722", "postDate": "12/15/2019 13:45:02", "content": "The score difference is not so meaningful, because you just try to overfit the public test data.
", "rawMarkdown": "The score difference is not so meaningful, because you just try to overfit the public test data.", "votes": null }, { "id": "695775", "postDate": "12/15/2019 14:42:50", "content": "this is truth, well done，thank you！！👍
", "rawMarkdown": "this is truth, well done，thank you！！👍", "votes": null }, { "id": "695979", "postDate": "12/16/2019 00:46:40", "content": "yeah I am having trouble pushing through .9890. think I am getting into the touch part were I need to tweak a hyper parameter just right and just get lucky lol
", "rawMarkdown": "yeah I am having trouble pushing through .9890. think I am getting into the touch part were I need to tweak a hyper parameter just right and just get lucky lol", "votes": null }, { "id": "696132", "postDate": "12/16/2019 06:58:05", "content": "good luck！！！ lol
", "rawMarkdown": "good luck！！！ lol", "votes": null }, { "id": "696473", "postDate": "12/16/2019 16:37:16", "content": "After a month of not breaking through 0.98900 I jumped to 0.99000 yesterday. I have tried so many different things including Logistic Regression on the output of 8 CNN models lol.
\n\nI went back to one of my simpler models and just increased my validation proportion to 0.25 from 0.20 and changed my random seed. I guess the new distribution of data was all it took.
", "rawMarkdown": "After a month of not breaking through 0.98900 I jumped to 0.99000 yesterday. I have tried so many different things including Logistic Regression on the output of 8 CNN models lol. \n\nI went back to one of my simpler models and just increased my validation proportion to 0.25 from 0.20 and changed my random seed. I guess the new distribution of data was all it took.", "votes": null }, { "id": "697114", "postDate": "12/17/2019 13:53:42", "content": "I have used swish, instead of other activation functions ( ReLu, LeakyReLu, Sigmoid etc.) So, I have increased my score 0.98 to 0.99
\n\nMore Info: https://www.machinecurve.com/index.php/2019/05/30/why-swish-could-perform-better-than-relu/
", "rawMarkdown": "I have used swish, instead of other activation functions ( ReLu, LeakyReLu, Sigmoid etc.) So, I have increased my score 0.98 to 0.99 \n\nMore Info: https://www.machinecurve.com/index.php/2019/05/30/why-swish-could-perform-better-than-relu/", "votes": null }, { "id": "697142", "postDate": "12/17/2019 14:28:09", "content": "Thanks a lot！I learned it from the website you shared. lol👍 I will\n try it right now.
", "rawMarkdown": "Thanks a lot！I learned it from the website you shared. lol👍 I will\n try it right now.", "votes": null }, { "id": "697487", "postDate": "12/18/2019 01:44:29", "content": "this is very useful information! thanks!
", "rawMarkdown": "this is very useful information! thanks!", "votes": null }, { "id": "697488", "postDate": "12/18/2019 01:52:05", "content": "Are you means, output with activation sigmoid function to 10 categories classifier on Dense layer?\nand ensemble it?
", "rawMarkdown": "Are you means, output with activation sigmoid function to 10 categories classifier on Dense layer?\nand ensemble it?", "votes": null } ], "comments": [ { "id": 695466, "author_name": "yyysjz", "author_url": "", "post_date": "12/15/2019 07:30:03", "content": "Now my highest score on LB is 0.98940....😫
", "votes": null, "replies": [ { "id": 695482, "author_name": "amitjslearn", "author_url": "", "post_date": "12/15/2019 08:09:42", "content": "Did your score change this time?
", "votes": null, "replies": [] }, { "id": 695505, "author_name": "yyysjz", "author_url": "", "post_date": "12/15/2019 09:01:42", "content": "nothing happened.😭
", "votes": null, "replies": [] } ] }, { "id": 695480, "author_name": "namanj27", "author_url": "", "post_date": "12/15/2019 08:05:04", "content": "I am sure that you must have tried all those fancy techniques( Earlystopping, reducing learning rate, ImageDataGenerator etc.) since your score is 0.989.\n I have experienced some kind of a weird thing happening here on kaggle, as I ran same code in two different times and got 0.988 in one and 0.992 in the other !\nSo I would suggest that you run your code again and see for these changing public scores for yourself.
", "votes": null, "replies": [ { "id": 695504, "author_name": "yyysjz", "author_url": "", "post_date": "12/15/2019 08:58:45", "content": "I also found that, the scores are different from different codes, I will try again, thank you!👍
", "votes": null, "replies": [] }, { "id": 695722, "author_name": "lhy1997", "author_url": "", "post_date": "12/15/2019 13:45:02", "content": "The score difference is not so meaningful, because you just try to overfit the public test data.
", "votes": null, "replies": [] }, { "id": 695775, "author_name": "yyysjz", "author_url": "", "post_date": "12/15/2019 14:42:50", "content": "this is truth, well done，thank you！！👍
", "votes": null, "replies": [] } ] }, { "id": 695979, "author_name": "georgedittmar", "author_url": "", "post_date": "12/16/2019 00:46:40", "content": "yeah I am having trouble pushing through .9890. think I am getting into the touch part were I need to tweak a hyper parameter just right and just get lucky lol
", "votes": null, "replies": [ { "id": 696132, "author_name": "yyysjz", "author_url": "", "post_date": "12/16/2019 06:58:05", "content": "good luck！！！ lol
", "votes": null, "replies": [] } ] }, { "id": 696473, "author_name": "mutantspore", "author_url": "", "post_date": "12/16/2019 16:37:16", "content": "After a month of not breaking through 0.98900 I jumped to 0.99000 yesterday. I have tried so many different things including Logistic Regression on the output of 8 CNN models lol.
\n\nI went back to one of my simpler models and just increased my validation proportion to 0.25 from 0.20 and changed my random seed. I guess the new distribution of data was all it took.
", "votes": null, "replies": [ { "id": 697488, "author_name": "ensembleman", "author_url": "", "post_date": "12/18/2019 01:52:05", "content": "Are you means, output with activation sigmoid function to 10 categories classifier on Dense layer?\nand ensemble it?
", "votes": null, "replies": [] } ] }, { "id": 697114, "author_name": "mcagriaksoy", "author_url": "", "post_date": "12/17/2019 13:53:42", "content": "I have used swish, instead of other activation functions ( ReLu, LeakyReLu, Sigmoid etc.) So, I have increased my score 0.98 to 0.99
\n\nMore Info: https://www.machinecurve.com/index.php/2019/05/30/why-swish-could-perform-better-than-relu/
", "votes": null, "replies": [ { "id": 697142, "author_name": "yyysjz", "author_url": "", "post_date": "12/17/2019 14:28:09", "content": "Thanks a lot！I learned it from the website you shared. lol👍 I will\n try it right now.
", "votes": null, "replies": [] }, { "id": 697487, "author_name": "ensembleman", "author_url": "", "post_date": "12/18/2019 01:44:29", "content": "this is very useful information! thanks!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "695465": "I've tried many, many times, and by any ways I can't reach the score of 0.99 or above on the LB. I would like to ask those teachers about 0.99 points, how does this work? thanks.🙏 🙏", "695466": "Now my highest score on LB is 0.98940....😫", "695480": "I am sure that you must have tried all those fancy techniques( Earlystopping, reducing learning rate, ImageDataGenerator etc.) since your score is 0.989.\n I have experienced some kind of a weird thing happening here on kaggle, as I ran same code in two different times and got 0.988 in one and 0.992 in the other !\nSo I would suggest that you run your code again and see for these changing public scores for yourself.", "695482": "Did your score change this time?", "695504": "I also found that, the scores are different from different codes, I will try again, thank you!👍", "695505": "nothing happened.😭", "695722": "The score difference is not so meaningful, because you just try to overfit the public test data.", "695775": "this is truth, well done，thank you！！👍", "695979": "yeah I am having trouble pushing through .9890. think I am getting into the touch part were I need to tweak a hyper parameter just right and just get lucky lol", "696132": "good luck！！！ lol", "696473": "After a month of not breaking through 0.98900 I jumped to 0.99000 yesterday. I have tried so many different things including Logistic Regression on the output of 8 CNN models lol. \n\nI went back to one of my simpler models and just increased my validation proportion to 0.25 from 0.20 and changed my random seed. I guess the new distribution of data was all it took.", "697114": "I have used swish, instead of other activation functions ( ReLu, LeakyReLu, Sigmoid etc.) So, I have increased my score 0.98 to 0.99 \n\nMore Info: https://www.machinecurve.com/index.php/2019/05/30/why-swish-could-perform-better-than-relu/", "697142": "Thanks a lot！I learned it from the website you shared. lol👍 I will\n try it right now.", "697487": "this is very useful information! thanks!", "697488": "Are you means, output with activation sigmoid function to 10 categories classifier on Dense layer?\nand ensemble it?" }, "source": "meta" }