Very impressive performance. Seems that the subject matter expertise for once really came through in a Kaggle competition. Can't wait to read your solution.
\n\nMany congratulations to all the other winners and participants in this competition. Wish I had more time to invest in it.
", "messages": [ { "id": "440733", "postDate": "12/18/2018 00:05:48", "content": "Very impressive performance. Seems that the subject matter expertise for once really came through in a Kaggle competition. Can't wait to read your solution.
\n\nMany congratulations to all the other winners and participants in this competition. Wish I had more time to invest in it.
", "rawMarkdown": "Very impressive performance. Seems that the subject matter expertise for once really came through in a Kaggle competition. Can't wait to read your solution.\n\nMany congratulations to all the other winners and participants in this competition. Wish I had more time to invest in it.", "votes": null }, { "id": "440748", "postDate": "12/18/2018 00:31:01", "content": "Thanks, and congratulations to everyone else! I learned a lot in this competition and I'll be interested to see what everybody did. Now to clean up my code and write everything up...
", "rawMarkdown": "Thanks, and congratulations to everyone else! I learned a lot in this competition and I'll be interested to see what everybody did. Now to clean up my code and write everything up...", "votes": null }, { "id": "440751", "postDate": "12/18/2018 00:33:47", "content": "Haha, good luck with that! ;)
", "rawMarkdown": "Haha, good luck with that! ;)", "votes": null }, { "id": "440781", "postDate": "12/18/2018 01:16:40", "content": "Congrats Kyle! Very well deserved. I believe astronomy knowledge helped a lot in this competition.
", "rawMarkdown": "Congrats Kyle! Very well deserved. I believe astronomy knowledge helped a lot in this competition.", "votes": null }, { "id": "440786", "postDate": "12/18/2018 01:19:50", "content": "Unfortunately, I never studied Astronomy, but have taught it for one semester. :)
", "rawMarkdown": "Unfortunately, I never studied Astronomy, but have taught it for one semester. :)", "votes": null }, { "id": "440815", "postDate": "12/18/2018 01:51:56", "content": "Congrats Kyle! Very much looking forward to seeing your solution.
", "rawMarkdown": "Congrats Kyle! Very much looking forward to seeing your solution.", "votes": null }, { "id": "440818", "postDate": "12/18/2018 01:55:39", "content": "Congratulations! Well deserved!
\n\nSpecial thank you very much for your comments and answers. It was quite altruistic and noble from your side to keep answering my questions. Thanks to you, I was able to read about k-correction, implement it and re-calculate features. (The k-corrected values did not add, maybe I've done it wrong, I used this calculator http://kcor.sai.msu.ru/getthecode/ and choosing colors suggested in the other paper with polynomial approximations of k-corr for z <0.5 https://arxiv.org/pdf/1410.8139.pdf ) However, the colors gave us a boost. I did not think to add them before I read those papers about colors in astronomy, so it was very helpful. Your last days comment about over-fitting also helped a lot. Before that, I was trying to find a black cat in a dark room. \n(*The hardest thing of all is to find a black cat in a dark room. Especially, if there is no cat. Confucius :) ).
\n\nIt was a very great help in the last few days, so thank you! :)
", "rawMarkdown": "Congratulations! Well deserved!\n\nSpecial thank you very much for your comments and answers. It was quite altruistic and noble from your side to keep answering my questions. Thanks to you, I was able to read about k-correction, implement it and re-calculate features. (The k-corrected values did not add, maybe I've done it wrong, I used this calculator http://kcor.sai.msu.ru/getthecode/ and choosing colors suggested in the other paper with polynomial approximations of k-corr for z <0.5 https://arxiv.org/pdf/1410.8139.pdf ) However, the colors gave us a boost. I did not think to add them before I read those papers about colors in astronomy, so it was very helpful. Your last days comment about over-fitting also helped a lot. Before that, I was trying to find a black cat in a dark room. \n(*The hardest thing of all is to find a black cat in a dark room. Especially, if there is no cat. Confucius :) ). \n\nIt was a very great help in the last few days, so thank you! :)", "votes": null }, { "id": "440826", "postDate": "12/18/2018 02:12:35", "content": "Astronomy knowledge appears to have helped less than you might imagine. We have some idea who other astronomers taking part in the challenge were from usernames/team names, and most are not in the running. Indeed, for more than half the contest, no astronomer was in the running. The scores are public in the raw data link -- there are some noticeable improvements right after when some kernels were made public, and others where a specific team made a big improvement, but outside of those jumps, it's just people gradually improving.
\n\nWe're really delighted with how many people participated and did really well, and again, there are other prizes for models that are useful even if you did not place on the top 3 of the leaderboard!
\n\nBest,
\n\n-Gautham for the PLAsTiCC team
", "rawMarkdown": "Astronomy knowledge appears to have helped less than you might imagine. We have some idea who other astronomers taking part in the challenge were from usernames/team names, and most are not in the running. Indeed, for more than half the contest, *no astronomer* was in the running. The scores are public in the raw data link -- there are some noticeable improvements right after when some kernels were made public, and others where a specific team made a big improvement, but outside of those jumps, it's just people gradually improving. \n\nWe're really delighted with how many people participated and did really well, and again, there are other prizes for models that are useful even if you did not place on the top 3 of the leaderboard!\n\nBest,\n\n-Gautham for the PLAsTiCC team", "votes": null }, { "id": "440832", "postDate": "12/18/2018 02:35:20", "content": "I actually didn't implement proper k corrections, so I can't comment on how helpful they would have been. My model does somewhat automatically capture parts of them. It is very difficult to do them right without having a spectral model of the objects in question! Getting that all right and understanding any biases that are introduced will be a big challenge for LSST...
", "rawMarkdown": "I actually didn't implement proper k corrections, so I can't comment on how helpful they would have been. My model does somewhat automatically capture parts of them. It is very difficult to do them right without having a spectral model of the objects in question! Getting that all right and understanding any biases that are introduced will be a big challenge for LSST...", "votes": null }, { "id": "440834", "postDate": "12/18/2018 02:38:54", "content": "I got into this competition so very late. Congratulations to Kyle, and a shout-out to some serious Astro representation in the Data Science community!
", "rawMarkdown": "I got into this competition so very late. Congratulations to Kyle, and a shout-out to some serious Astro representation in the Data Science community!", "votes": null }, { "id": "441240", "postDate": "12/18/2018 12:59:03", "content": "Bojan,
\n\nWhere is your write up? ;)
", "rawMarkdown": "Bojan,\n\nWhere is your write up? ;)", "votes": null }, { "id": "441300", "postDate": "12/18/2018 14:13:50", "content": "We’ll unfortunately probably refrain from writing anything up. For the most part our solution was based on very straightforward use of the “standard” modeling tricks - LIghtGBM, XGBoost, etcc., and you can see some of those in Olivier’s kernels. There are a few tricks that we used that we’ll keep to ourselves for now. I have one interesting insight that I hope to share in the kernels, provided that it can fit in that space.
", "rawMarkdown": "We’ll unfortunately probably refrain from writing anything up. For the most part our solution was based on very straightforward use of the “standard” modeling tricks - LIghtGBM, XGBoost, etcc., and you can see some of those in Olivier’s kernels. There are a few tricks that we used that we’ll keep to ourselves for now. I have one interesting insight that I hope to share in the kernels, provided that it can fit in that space.", "votes": null }, { "id": "441302", "postDate": "12/18/2018 14:16:54", "content": "\n\n\nThere are a few tricks that we used that we’ll keep to ourselves for now.
\n
Too bad.
", "rawMarkdown": "> There are a few tricks that we used that we’ll keep to ourselves for now.\n\nToo bad.", "votes": null } ], "comments": [ { "id": 440748, "author_name": "kyleboone", "author_url": "", "post_date": "12/18/2018 00:31:01", "content": "Thanks, and congratulations to everyone else! I learned a lot in this competition and I'll be interested to see what everybody did. Now to clean up my code and write everything up...
", "votes": null, "replies": [ { "id": 440751, "author_name": "tunguz", "author_url": "", "post_date": "12/18/2018 00:33:47", "content": "Haha, good luck with that! ;)
", "votes": null, "replies": [] } ] }, { "id": 440781, "author_name": "titericz", "author_url": "", "post_date": "12/18/2018 01:16:40", "content": "Congrats Kyle! Very well deserved. I believe astronomy knowledge helped a lot in this competition.
", "votes": null, "replies": [ { "id": 440786, "author_name": "tunguz", "author_url": "", "post_date": "12/18/2018 01:19:50", "content": "Unfortunately, I never studied Astronomy, but have taught it for one semester. :)
", "votes": null, "replies": [] }, { "id": 440826, "author_name": "gsnarayan", "author_url": "", "post_date": "12/18/2018 02:12:35", "content": "Astronomy knowledge appears to have helped less than you might imagine. We have some idea who other astronomers taking part in the challenge were from usernames/team names, and most are not in the running. Indeed, for more than half the contest, no astronomer was in the running. The scores are public in the raw data link -- there are some noticeable improvements right after when some kernels were made public, and others where a specific team made a big improvement, but outside of those jumps, it's just people gradually improving.
\n\nWe're really delighted with how many people participated and did really well, and again, there are other prizes for models that are useful even if you did not place on the top 3 of the leaderboard!
\n\nBest,
\n\n-Gautham for the PLAsTiCC team
", "votes": null, "replies": [] } ] }, { "id": 440815, "author_name": "jackvial", "author_url": "", "post_date": "12/18/2018 01:51:56", "content": "Congrats Kyle! Very much looking forward to seeing your solution.
", "votes": null, "replies": [] }, { "id": 440818, "author_name": "blondinka", "author_url": "", "post_date": "12/18/2018 01:55:39", "content": "Congratulations! Well deserved!
\n\nSpecial thank you very much for your comments and answers. It was quite altruistic and noble from your side to keep answering my questions. Thanks to you, I was able to read about k-correction, implement it and re-calculate features. (The k-corrected values did not add, maybe I've done it wrong, I used this calculator http://kcor.sai.msu.ru/getthecode/ and choosing colors suggested in the other paper with polynomial approximations of k-corr for z <0.5 https://arxiv.org/pdf/1410.8139.pdf ) However, the colors gave us a boost. I did not think to add them before I read those papers about colors in astronomy, so it was very helpful. Your last days comment about over-fitting also helped a lot. Before that, I was trying to find a black cat in a dark room. \n(*The hardest thing of all is to find a black cat in a dark room. Especially, if there is no cat. Confucius :) ).
\n\nIt was a very great help in the last few days, so thank you! :)
", "votes": null, "replies": [ { "id": 440832, "author_name": "kyleboone", "author_url": "", "post_date": "12/18/2018 02:35:20", "content": "I actually didn't implement proper k corrections, so I can't comment on how helpful they would have been. My model does somewhat automatically capture parts of them. It is very difficult to do them right without having a spectral model of the objects in question! Getting that all right and understanding any biases that are introduced will be a big challenge for LSST...
", "votes": null, "replies": [] } ] }, { "id": 440834, "author_name": "delayedkarma", "author_url": "", "post_date": "12/18/2018 02:38:54", "content": "I got into this competition so very late. Congratulations to Kyle, and a shout-out to some serious Astro representation in the Data Science community!
", "votes": null, "replies": [] }, { "id": 441240, "author_name": "cpmpml", "author_url": "", "post_date": "12/18/2018 12:59:03", "content": "Bojan,
\n\nWhere is your write up? ;)
", "votes": null, "replies": [ { "id": 441300, "author_name": "tunguz", "author_url": "", "post_date": "12/18/2018 14:13:50", "content": "We’ll unfortunately probably refrain from writing anything up. For the most part our solution was based on very straightforward use of the “standard” modeling tricks - LIghtGBM, XGBoost, etcc., and you can see some of those in Olivier’s kernels. There are a few tricks that we used that we’ll keep to ourselves for now. I have one interesting insight that I hope to share in the kernels, provided that it can fit in that space.
", "votes": null, "replies": [] }, { "id": 441302, "author_name": "cpmpml", "author_url": "", "post_date": "12/18/2018 14:16:54", "content": "\n\n\nThere are a few tricks that we used that we’ll keep to ourselves for now.
\n
Too bad.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "440733": "Very impressive performance. Seems that the subject matter expertise for once really came through in a Kaggle competition. Can't wait to read your solution.\n\nMany congratulations to all the other winners and participants in this competition. Wish I had more time to invest in it.", "440748": "Thanks, and congratulations to everyone else! I learned a lot in this competition and I'll be interested to see what everybody did. Now to clean up my code and write everything up...", "440751": "Haha, good luck with that! ;)", "440781": "Congrats Kyle! Very well deserved. I believe astronomy knowledge helped a lot in this competition.", "440786": "Unfortunately, I never studied Astronomy, but have taught it for one semester. :)", "440815": "Congrats Kyle! Very much looking forward to seeing your solution.", "440818": "Congratulations! Well deserved!\n\nSpecial thank you very much for your comments and answers. It was quite altruistic and noble from your side to keep answering my questions. Thanks to you, I was able to read about k-correction, implement it and re-calculate features. (The k-corrected values did not add, maybe I've done it wrong, I used this calculator http://kcor.sai.msu.ru/getthecode/ and choosing colors suggested in the other paper with polynomial approximations of k-corr for z <0.5 https://arxiv.org/pdf/1410.8139.pdf ) However, the colors gave us a boost. I did not think to add them before I read those papers about colors in astronomy, so it was very helpful. Your last days comment about over-fitting also helped a lot. Before that, I was trying to find a black cat in a dark room. \n(*The hardest thing of all is to find a black cat in a dark room. Especially, if there is no cat. Confucius :) ). \n\nIt was a very great help in the last few days, so thank you! :)", "440826": "Astronomy knowledge appears to have helped less than you might imagine. We have some idea who other astronomers taking part in the challenge were from usernames/team names, and most are not in the running. Indeed, for more than half the contest, *no astronomer* was in the running. The scores are public in the raw data link -- there are some noticeable improvements right after when some kernels were made public, and others where a specific team made a big improvement, but outside of those jumps, it's just people gradually improving. \n\nWe're really delighted with how many people participated and did really well, and again, there are other prizes for models that are useful even if you did not place on the top 3 of the leaderboard!\n\nBest,\n\n-Gautham for the PLAsTiCC team", "440832": "I actually didn't implement proper k corrections, so I can't comment on how helpful they would have been. My model does somewhat automatically capture parts of them. It is very difficult to do them right without having a spectral model of the objects in question! Getting that all right and understanding any biases that are introduced will be a big challenge for LSST...", "440834": "I got into this competition so very late. Congratulations to Kyle, and a shout-out to some serious Astro representation in the Data Science community!", "441240": "Bojan,\n\nWhere is your write up? ;)", "441300": "We’ll unfortunately probably refrain from writing anything up. For the most part our solution was based on very straightforward use of the “standard” modeling tricks - LIghtGBM, XGBoost, etcc., and you can see some of those in Olivier’s kernels. There are a few tricks that we used that we’ll keep to ourselves for now. I have one interesting insight that I hope to share in the kernels, provided that it can fit in that space.", "441302": "> There are a few tricks that we used that we’ll keep to ourselves for now.\n\nToo bad." }, "source": "meta" }