Congrats also to all gold, silver and bronze medal teams! And specially the ones how made it solo: dhammack, JohnJayChou, tkuanlun350 and Igor Praznik. This is also a HUGE achievement.\nDouble congrats to @dhammack to become a GM !
\n\nA special congrats to @bestfitting who won solo and top rank two deep learning competitions at the same same. This is not for regular people. Wondering how many GPUs, resources and spare time do you have to compete? ;-)
\n\nGiba
", "messages": [ { "id": "421391", "postDate": "11/15/2018 00:33:19", "content": "Congrats also to all gold, silver and bronze medal teams! And specially the ones how made it solo: dhammack, JohnJayChou, tkuanlun350 and Igor Praznik. This is also a HUGE achievement.\nDouble congrats to @dhammack to become a GM !
\n\nA special congrats to @bestfitting who won solo and top rank two deep learning competitions at the same same. This is not for regular people. Wondering how many GPUs, resources and spare time do you have to compete? ;-)
\n\nGiba
", "rawMarkdown": "# Congratulations to all winners: \n#1 - Konstantin Gavrilchik and Evgeny Kononenko\n#2 - Victor Durnov and Selim Seferbekov\n#3 - bestfitting\n\nCongrats also to all gold, silver and bronze medal teams! And specially the ones how made it solo: dhammack, JohnJayChou, tkuanlun350 and Igor Praznik. This is also a HUGE achievement.\nDouble congrats to @dhammack to become a GM !\n\nA special congrats to @bestfitting who won solo and top rank two deep learning competitions at the same same. This is not for regular people. Wondering how many GPUs, resources and spare time do you have to compete? ;-)\n\nGiba", "votes": null }, { "id": "421504", "postDate": "11/15/2018 04:38:28", "content": "Congratulations to the winners.
", "rawMarkdown": "Congratulations to the winners.", "votes": null }, { "id": "421542", "postDate": "11/15/2018 05:35:47", "content": "Congratulations to the winners, after looking at private Leaderboard my first thought was to go and register to ods.ai :)
", "rawMarkdown": "Congratulations to the winners, after looking at private Leaderboard my first thought was to go and register to ods.ai :)", "votes": null }, { "id": "421640", "postDate": "11/15/2018 08:17:38", "content": "bestfitting has gotten gold in almost every competition, no matter how complex, how unconventional they were. I don't know many people in this world who can do that, especially when they have their day jobs as well. Either he's incredibly brilliant, or......something. I'm curious to know about the arcane magic.
", "rawMarkdown": "bestfitting has gotten gold in almost every competition, no matter how complex, how unconventional they were. I don't know many people in this world who can do that, especially when they have their day jobs as well. Either he's incredibly brilliant, or......something. I'm curious to know about the arcane magic.", "votes": null }, { "id": "421670", "postDate": "11/15/2018 08:46:21", "content": "Congratulations Winners
", "rawMarkdown": "Congratulations Winners", "votes": null }, { "id": "421790", "postDate": "11/15/2018 12:27:07", "content": "Congratulations to all the winners.
", "rawMarkdown": "Congratulations to all the winners.", "votes": null }, { "id": "421803", "postDate": "11/15/2018 12:50:28", "content": "I think i know how he does it. Ever seen the move \"the prestige\"??
", "rawMarkdown": "I think i know how he does it. Ever seen the move \"the prestige\"??", "votes": null }, { "id": "421832", "postDate": "11/15/2018 13:27:22", "content": "Not yet. Maybe I should.
", "rawMarkdown": "Not yet. Maybe I should.", "votes": null }, { "id": "421853", "postDate": "11/15/2018 13:51:52", "content": "I'm currently doing the mlcourse.ai MOOC, which I think is part of ODS.ai. I have learned a lot so far. I'm excited for when I'm done with the course and have more time to apply my skills to kaggle competitions.
", "rawMarkdown": "I'm currently doing the mlcourse.ai MOOC, which I think is part of ODS.ai. I have learned a lot so far. I'm excited for when I'm done with the course and have more time to apply my skills to kaggle competitions.", "votes": null }, { "id": "422401", "postDate": "11/16/2018 07:09:20", "content": "Congratulations winners.
", "rawMarkdown": "Congratulations winners.", "votes": null }, { "id": "422571", "postDate": "11/16/2018 12:26:09", "content": "@dhammack makes sense ;-)
", "rawMarkdown": "dhammack makes sense ;-)", "votes": null }, { "id": "422640", "postDate": "11/16/2018 14:54:39", "content": "Congrats
", "rawMarkdown": "Congrats", "votes": null }, { "id": "422787", "postDate": "11/16/2018 19:24:43", "content": "Thank you!\nIt's quite lucky to get top 3 in this competition.\nThere are so many great and kind kagglers at kaggle,I am only a common one,there is no magic.\nAccording to my limited experiences on kaggle competitions,we all want to do more experiments during a limited time,there are a lot of ways:finding some teamates,getting more computation resources,and investing more time...\nIn my humble opinion,the most important factor is efficiency.\nTo improve efficiency,learning from others and learn from previous competitions are important.\nFor example,my solution of this competition is quite similar to SeaLion competition,only replace the UNET model to SALT competition's.To save training time,I cropped 256x256 patchs where may contain ships from 768x768 images, just as I had done in sealion competition,so I can train models on a 1080Ti GPU easily\nAnd I try to do experiments on small and simple model and data as I did in CDiscount competition,it is very important in a competition with large dataset,I find res18 with 96x96 size input is enough for experiments in Draw competition,I am training model on my old machine with 4 Titan Maxwell GPUs :)\nLearning from others(absolutley including you @titericz) is very helpful to me,during SALT and this competition,I read all the solutions of related kaggle competitions,for example,DSB 2018(and I read your @dhammack solution of DSB2017 solution carefully after I entered Carvana Competition) so I can make sure I did not miss any skills and tricks related.By doing so,I can 'Teamup' with all the winners.\nAs we know,the deep learning is very hot in both academic and industry,so I read papers everyday,ICCV,CVPR....,and search and read all related papers during a competition(>100 per competition often),what's more, I also read source codes of them or reproduce some of them(I find if we can be proficient in using Keras,pytorch,caffe,tensorflow,it will be more efficient,my experiences as an enthusiastic programmer helped me a lot,I read, modified a lot of software and built solutions on them,such as Android,Eclipse,Birt,Hadoop,Mahout,MySql...these names bring back so many bitter-sweet memories )When I need them in a competition,I can select some most promising one,for example,I find CBAM is a very good attention method,so I used it in SALT and this competition,and it's better than SE mechamism.By reading papers,we can 'Teamup' with most professional persons in the world in a certain field.I think these are also @hengck23 and @mihaskalic 's everyday pratices.\nAs to my everyday life and job,I do physical exercises every day at least for an hour and I can arrange my work freely,so I can start a training and predicting batch and then launch a telephone meeting to discuss some projects in real life,I think they are much easier than kaggle and algorithms such as BERT--I'm reading now :)\n
Awesome!!
", "rawMarkdown": "Awesome!!", "votes": null }, { "id": "423166", "postDate": "11/17/2018 16:54:15", "content": "Congratulations everybody and thanks to kagglers for sharing and help me to learn a little more
", "rawMarkdown": "Congratulations everybody and thanks to kagglers for sharing and help me to learn a little more", "votes": null }, { "id": "423399", "postDate": "11/18/2018 06:40:17", "content": "Congratulations
", "rawMarkdown": "Congratulations", "votes": null }, { "id": "423417", "postDate": "11/18/2018 07:43:44", "content": "congratulations
", "rawMarkdown": "congratulations", "votes": null }, { "id": "423519", "postDate": "11/18/2018 13:27:53", "content": "Congratulations winners.
", "rawMarkdown": "Congratulations winners.", "votes": null }, { "id": "424375", "postDate": "11/20/2018 02:22:55", "content": "Given that this was my first competition to participate in I am utterly overwhelmed my the positiveness and openness of this community! Congratulations to the winners and to all the other hard working participants. Thank you all for sharing your knowledge and experiences (special thanks to @Iafoss for the outstanding kernels).
", "rawMarkdown": "Given that this was my first competition to participate in I am utterly overwhelmed my the positiveness and openness of this community! Congratulations to the winners and to all the other hard working participants. Thank you all for sharing your knowledge and experiences (special thanks to @Iafoss for the outstanding kernels).", "votes": null }, { "id": "424509", "postDate": "11/20/2018 08:22:42", "content": "Congratulations to the winners!
", "rawMarkdown": "Congratulations to the winners!", "votes": null }, { "id": "424651", "postDate": "11/20/2018 13:44:04", "content": "congratulations
", "rawMarkdown": "congratulations", "votes": null }, { "id": "456798", "postDate": "01/16/2019 15:06:24", "content": "can you share your code?
", "rawMarkdown": "can you share your code?", "votes": null } ], "comments": [ { "id": 421504, "author_name": "sheriytm", "author_url": "", "post_date": "11/15/2018 04:38:28", "content": "Congratulations to the winners.
", "votes": null, "replies": [] }, { "id": 421542, "author_name": "demonplus", "author_url": "", "post_date": "11/15/2018 05:35:47", "content": "Congratulations to the winners, after looking at private Leaderboard my first thought was to go and register to ods.ai :)
", "votes": null, "replies": [ { "id": 421853, "author_name": "afgonczol", "author_url": "", "post_date": "11/15/2018 13:51:52", "content": "I'm currently doing the mlcourse.ai MOOC, which I think is part of ODS.ai. I have learned a lot so far. I'm excited for when I'm done with the course and have more time to apply my skills to kaggle competitions.
", "votes": null, "replies": [] } ] }, { "id": 421640, "author_name": "nzabcd", "author_url": "", "post_date": "11/15/2018 08:17:38", "content": "bestfitting has gotten gold in almost every competition, no matter how complex, how unconventional they were. I don't know many people in this world who can do that, especially when they have their day jobs as well. Either he's incredibly brilliant, or......something. I'm curious to know about the arcane magic.
", "votes": null, "replies": [ { "id": 421803, "author_name": "dhammack", "author_url": "", "post_date": "11/15/2018 12:50:28", "content": "I think i know how he does it. Ever seen the move \"the prestige\"??
", "votes": null, "replies": [] }, { "id": 421832, "author_name": "nzabcd", "author_url": "", "post_date": "11/15/2018 13:27:22", "content": "Not yet. Maybe I should.
", "votes": null, "replies": [] }, { "id": 422571, "author_name": "titericz", "author_url": "", "post_date": "11/16/2018 12:26:09", "content": "@dhammack makes sense ;-)
", "votes": null, "replies": [] } ] }, { "id": 421670, "author_name": "", "author_url": "", "post_date": "11/15/2018 08:46:21", "content": "Congratulations Winners
", "votes": null, "replies": [] }, { "id": 421790, "author_name": "nirajtiwary", "author_url": "", "post_date": "11/15/2018 12:27:07", "content": "Congratulations to all the winners.
", "votes": null, "replies": [] }, { "id": 422401, "author_name": "vj1998", "author_url": "", "post_date": "11/16/2018 07:09:20", "content": "Congratulations winners.
", "votes": null, "replies": [] }, { "id": 422640, "author_name": "", "author_url": "", "post_date": "11/16/2018 14:54:39", "content": "Congrats
", "votes": null, "replies": [] }, { "id": 422787, "author_name": "bestfitting", "author_url": "", "post_date": "11/16/2018 19:24:43", "content": "Thank you!\nIt's quite lucky to get top 3 in this competition.\nThere are so many great and kind kagglers at kaggle,I am only a common one,there is no magic.\nAccording to my limited experiences on kaggle competitions,we all want to do more experiments during a limited time,there are a lot of ways:finding some teamates,getting more computation resources,and investing more time...\nIn my humble opinion,the most important factor is efficiency.\nTo improve efficiency,learning from others and learn from previous competitions are important.\nFor example,my solution of this competition is quite similar to SeaLion competition,only replace the UNET model to SALT competition's.To save training time,I cropped 256x256 patchs where may contain ships from 768x768 images, just as I had done in sealion competition,so I can train models on a 1080Ti GPU easily\nAnd I try to do experiments on small and simple model and data as I did in CDiscount competition,it is very important in a competition with large dataset,I find res18 with 96x96 size input is enough for experiments in Draw competition,I am training model on my old machine with 4 Titan Maxwell GPUs :)\nLearning from others(absolutley including you @titericz) is very helpful to me,during SALT and this competition,I read all the solutions of related kaggle competitions,for example,DSB 2018(and I read your @dhammack solution of DSB2017 solution carefully after I entered Carvana Competition) so I can make sure I did not miss any skills and tricks related.By doing so,I can 'Teamup' with all the winners.\nAs we know,the deep learning is very hot in both academic and industry,so I read papers everyday,ICCV,CVPR....,and search and read all related papers during a competition(>100 per competition often),what's more, I also read source codes of them or reproduce some of them(I find if we can be proficient in using Keras,pytorch,caffe,tensorflow,it will be more efficient,my experiences as an enthusiastic programmer helped me a lot,I read, modified a lot of software and built solutions on them,such as Android,Eclipse,Birt,Hadoop,Mahout,MySql...these names bring back so many bitter-sweet memories )When I need them in a competition,I can select some most promising one,for example,I find CBAM is a very good attention method,so I used it in SALT and this competition,and it's better than SE mechamism.By reading papers,we can 'Teamup' with most professional persons in the world in a certain field.I think these are also @hengck23 and @mihaskalic 's everyday pratices.\nAs to my everyday life and job,I do physical exercises every day at least for an hour and I can arrange my work freely,so I can start a training and predicting batch and then launch a telephone meeting to discuss some projects in real life,I think they are much easier than kaggle and algorithms such as BERT--I'm reading now :)\n
Awesome!!
", "votes": null, "replies": [] } ] }, { "id": 423166, "author_name": "smrucv", "author_url": "", "post_date": "11/17/2018 16:54:15", "content": "Congratulations everybody and thanks to kagglers for sharing and help me to learn a little more
", "votes": null, "replies": [] }, { "id": 423399, "author_name": "sailendrad", "author_url": "", "post_date": "11/18/2018 06:40:17", "content": "Congratulations
", "votes": null, "replies": [] }, { "id": 423417, "author_name": "wangbo3000", "author_url": "", "post_date": "11/18/2018 07:43:44", "content": "congratulations
", "votes": null, "replies": [] }, { "id": 423519, "author_name": "mprudentes", "author_url": "", "post_date": "11/18/2018 13:27:53", "content": "Congratulations winners.
", "votes": null, "replies": [] }, { "id": 424375, "author_name": "blueview", "author_url": "", "post_date": "11/20/2018 02:22:55", "content": "Given that this was my first competition to participate in I am utterly overwhelmed my the positiveness and openness of this community! Congratulations to the winners and to all the other hard working participants. Thank you all for sharing your knowledge and experiences (special thanks to @Iafoss for the outstanding kernels).
", "votes": null, "replies": [] }, { "id": 424509, "author_name": "scullwm", "author_url": "", "post_date": "11/20/2018 08:22:42", "content": "Congratulations to the winners!
", "votes": null, "replies": [] }, { "id": 424651, "author_name": "theworlddata", "author_url": "", "post_date": "11/20/2018 13:44:04", "content": "congratulations
", "votes": null, "replies": [] }, { "id": 456798, "author_name": "xiaojidan", "author_url": "", "post_date": "01/16/2019 15:06:24", "content": "can you share your code?
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "421391": "# Congratulations to all winners: \n#1 - Konstantin Gavrilchik and Evgeny Kononenko\n#2 - Victor Durnov and Selim Seferbekov\n#3 - bestfitting\n\nCongrats also to all gold, silver and bronze medal teams! And specially the ones how made it solo: dhammack, JohnJayChou, tkuanlun350 and Igor Praznik. This is also a HUGE achievement.\nDouble congrats to @dhammack to become a GM !\n\nA special congrats to @bestfitting who won solo and top rank two deep learning competitions at the same same. This is not for regular people. Wondering how many GPUs, resources and spare time do you have to compete? ;-)\n\nGiba", "421504": "Congratulations to the winners.", "421542": "Congratulations to the winners, after looking at private Leaderboard my first thought was to go and register to ods.ai :)", "421640": "bestfitting has gotten gold in almost every competition, no matter how complex, how unconventional they were. I don't know many people in this world who can do that, especially when they have their day jobs as well. Either he's incredibly brilliant, or......something. I'm curious to know about the arcane magic.", "421670": "Congratulations Winners", "421790": "Congratulations to all the winners.", "421803": "I think i know how he does it. Ever seen the move \"the prestige\"??", "421832": "Not yet. Maybe I should.", "421853": "I'm currently doing the mlcourse.ai MOOC, which I think is part of ODS.ai. I have learned a lot so far. I'm excited for when I'm done with the course and have more time to apply my skills to kaggle competitions.", "422401": "Congratulations winners.", "422571": "dhammack makes sense ;-)", "422640": "Congrats", "422787": "Thank you!\nIt's quite lucky to get top 3 in this competition.\nThere are so many great and kind kagglers at kaggle,I am only a common one,there is no magic.\nAccording to my limited experiences on kaggle competitions,we all want to do more experiments during a limited time,there are a lot of ways:finding some teamates,getting more computation resources,and investing more time...\nIn my humble opinion,the most important factor is efficiency.\nTo improve efficiency,learning from others and learn from previous competitions are important.\nFor example,my solution of this competition is quite similar to SeaLion competition,only replace the UNET model to SALT competition's.To save training time,I cropped 256x256 patchs where may contain ships from 768x768 images, just as I had done in sealion competition,so I can train models on a 1080Ti GPU easily\nAnd I try to do experiments on small and simple model and data as I did in CDiscount competition,it is very important in a competition with large dataset,I find res18 with 96x96 size input is enough for experiments in Draw competition,I am training model on my old machine with 4 Titan Maxwell GPUs :)\nLearning from others(absolutley including you @titericz) is very helpful to me,during SALT and this competition,I read all the solutions of related kaggle competitions,for example,DSB 2018(and I read your @dhammack solution of DSB2017 solution carefully after I entered Carvana Competition) so I can make sure I did not miss any skills and tricks related.By doing so,I can 'Teamup' with all the winners.\nAs we know,the deep learning is very hot in both academic and industry,so I read papers everyday,ICCV,CVPR....,and search and read all related papers during a competition(>100 per competition often),what's more, I also read source codes of them or reproduce some of them(I find if we can be proficient in using Keras,pytorch,caffe,tensorflow,it will be more efficient,my experiences as an enthusiastic programmer helped me a lot,I read, modified a lot of software and built solutions on them,such as Android,Eclipse,Birt,Hadoop,Mahout,MySql...these names bring back so many bitter-sweet memories )When I need them in a competition,I can select some most promising one,for example,I find CBAM is a very good attention method,so I used it in SALT and this competition,and it's better than SE mechamism.By reading papers,we can 'Teamup' with most professional persons in the world in a certain field.I think these are also @hengck23 and @mihaskalic 's everyday pratices.\nAs to my everyday life and job,I do physical exercises every day at least for an hour and I can arrange my work freely,so I can start a training and predicting batch and then launch a telephone meeting to discuss some projects in real life,I think they are much easier than kaggle and algorithms such as BERT--I'm reading now :)\n