{ "id": 16966, "title": "Magic hint to reduce CVM?", "url": "/competitions/flavours-of-physics/discussion/16966", "author_name": "", "post_date": "2015-10-12T17:11:45.623Z", "votes": null, "comment_count": 10, "views": 773, "content": "

Has somebody - who want to share - some magic hint to reduce CVM?

", "messages": [ { "id": "95854", "postDate": "10/12/2015 17:11:45", "content": "

Has somebody - who want to share - some magic hint to reduce CVM?

", "rawMarkdown": "Has somebody - who want to share - some magic hint to reduce CVM?", "votes": null }, { "id": "95859", "postDate": "10/12/2015 18:06:24", "content": "

[quote=George Solymosi;95854]

\n\n

Has somebody - who want to share - some magic hint to reduce CVM?

\n\n

[/quote]\nmy hint for never worrying about CvM test: try removing features correlated with tau invariant mass. CvM test exams if the model depends on tau mass or not, because the control channel is Ds decay which has a small mass difference.

\n\n

Using the current information, you might reconstruct the tau invariant mass (or something correlated with this). I tested it, and it worked nice with +0.003 to +0.005 AUC at least, but these features may not have physical meanings which are useless for the real life, so I choice not to have these features.

", "rawMarkdown": "[quote=George Solymosi;95854]\r\n\r\nHas somebody - who want to share - some magic hint to reduce CVM? \r\n\r\n[/quote]\r\nmy hint for never worrying about CvM test: try removing features correlated with tau invariant mass. CvM test exams if the model depends on tau mass or not, because the control channel is Ds decay which has a small mass difference. \r\n\r\nUsing the current information, you **might** reconstruct the tau invariant mass (or something correlated with this). I tested it, and it worked nice with +0.003 to +0.005 AUC at least, but these features may not have physical meanings which are useless for the real life, so I choice not to have these features.", "votes": null }, { "id": "95872", "postDate": "10/12/2015 19:04:59", "content": "

@phunter Thank you very much again p! The question is - at my current physics level - which are these? :)

", "rawMarkdown": "phunter Thank you very much again p! The question is - at my current physics level - which are these? :)", "votes": null }, { "id": "95875", "postDate": "10/12/2015 19:18:31", "content": "

[quote=George Solymosi;95872]

\n\n

@phunter Thank you very much again p! The question is - at my current physics level - which are these? :)

\n\n

[/quote]\nfeatures with pt and p correlations may imply some mass correlation, like tau pt vs the momentum of the other 3 particles, I guess.

", "rawMarkdown": "[quote=George Solymosi;95872]\r\n\r\n@phunter Thank you very much again p! The question is - at my current physics level - which are these? :)\r\n\r\n[/quote]\r\nfeatures with pt and p correlations may imply some mass correlation, like tau pt vs the momentum of the other 3 particles, I guess.", "votes": null }, { "id": "95880", "postDate": "10/12/2015 19:36:07", "content": "

Thank you I thought about pt and dira.

", "rawMarkdown": "Thank you I thought about pt and dira.", "votes": null }, { "id": "95883", "postDate": "10/12/2015 20:08:46", "content": "

[quote=George Solymosi;95880]

\n\n

Thank you I thought about pt and dira.

\n\n

[/quote]\npt and dira is fine, no problem, but pt with dira vs p0/p2/p3 (pt0/pt1/pt2) can give some mass correlation. Let's think of energy conservation where a flying particle decays into 3 other flying particles where we know the mass of the 3 one (they are muons).

", "rawMarkdown": "[quote=George Solymosi;95880]\r\n\r\nThank you I thought about pt and dira.\r\n\r\n[/quote]\r\npt and dira is fine, no problem, but pt with dira vs p0/p2/p3 (pt0/pt1/pt2) can give some mass correlation. Let's think of energy conservation where a flying particle decays into 3 other flying particles where we know the mass of the 3 one (they are muons).", "votes": null }, { "id": "95894", "postDate": "10/12/2015 21:04:57", "content": "

Thank you again, the key achievements of this (far not flawless) competition, that I learnt a lot, my things started to work at last and "met" great guys and scientists like you, Sir.

", "rawMarkdown": "Thank you again, the key achievements of this (far not flawless) competition, that I learnt a lot, my things started to work at last and \"met\" great guys and scientists like you, Sir.", "votes": null }, { "id": "95896", "postDate": "10/12/2015 21:09:42", "content": "

[quote=George Solymosi;95894]

\n\n

Thank you again, the key achievements of this (far not flawless) competition, that I learnt a lot, my things started to work at last and "met" great guys and scientists like you, Sir.

\n\n

[/quote]\nYou are so welcomed, and your score beats my score

", "rawMarkdown": "[quote=George Solymosi;95894]\r\n\r\nThank you again, the key achievements of this (far not flawless) competition, that I learnt a lot, my things started to work at last and \"met\" great guys and scientists like you, Sir.\r\n\r\n[/quote]\r\nYou are so welcomed, and your score beats my score", "votes": null }, { "id": "95897", "postDate": "10/12/2015 21:18:39", "content": "

As I mentioned before my score is just an experiment of mine with some obvious trick, thus at the end of all I am sure your score will much more "physically sound" and probably on LB will be higher as well. ;)

", "rawMarkdown": "As I mentioned before my score is just an experiment of mine with some obvious trick, thus at the end of all I am sure your score will much more \"physically sound\" and probably on LB will be higher as well. ;)", "votes": null }, { "id": "95902", "postDate": "10/12/2015 22:11:30", "content": "

[quote=phunter;95896]\nYou are so welcomed, and your score beats my score\n[/quote]

\n\n

UPDATE: I have a physically sound (meaning: no special trick) 0.990149 already! :) I wish I had more time.

", "rawMarkdown": "[quote=phunter;95896]\r\nYou are so welcomed, and your score beats my score\r\n[/quote]\r\n\r\n**UPDATE:** I have a physically sound (meaning: no special trick) **0.990149** already! :) I wish I had more time.", "votes": null }, { "id": "95903", "postDate": "10/12/2015 22:16:12", "content": "

[quote=George Solymosi;95902]

\n\n

[quote=phunter;95896]\nYou are so welcomed, and your score beats my score\n[/quote]

\n\n

UPDATE: I have a physically sound 0.990149 already! :) I wish I had more time.

\n\n

[/quote]\nWould be very nice to read your solution after the competition, and meaningful physics features are interesting.

\n\n

I learned in this competition that UGrad was nice, but I knew it too later and couldn't have time for coding a grid search interface for searching the parameters. So my solution was only xgboost + RF which may not look good finally: I just like xgboost so much. I will fix the pip installation for xgboost after this competition.

", "rawMarkdown": "[quote=George Solymosi;95902]\r\n\r\n[quote=phunter;95896]\r\nYou are so welcomed, and your score beats my score\r\n[/quote]\r\n\r\n**UPDATE:** I have a physically sound **0.990149** already! :) I wish I had more time.\r\n\r\n\r\n[/quote]\r\nWould be very nice to read your solution after the competition, and meaningful physics features are interesting. \r\n\r\nI learned in this competition that UGrad was nice, but I knew it too later and couldn't have time for coding a grid search interface for searching the parameters. So my solution was only xgboost + RF which may not look good finally: I just like xgboost so much. I will fix the pip installation for xgboost after this competition.", "votes": null } ], "comments": [ { "id": 95859, "author_name": "phunter", "author_url": "", "post_date": "10/12/2015 18:06:24", "content": "

[quote=George Solymosi;95854]

\n\n

Has somebody - who want to share - some magic hint to reduce CVM?

\n\n

[/quote]\nmy hint for never worrying about CvM test: try removing features correlated with tau invariant mass. CvM test exams if the model depends on tau mass or not, because the control channel is Ds decay which has a small mass difference.

\n\n

Using the current information, you might reconstruct the tau invariant mass (or something correlated with this). I tested it, and it worked nice with +0.003 to +0.005 AUC at least, but these features may not have physical meanings which are useless for the real life, so I choice not to have these features.

", "votes": null, "replies": [] }, { "id": 95872, "author_name": "", "author_url": "", "post_date": "10/12/2015 19:04:59", "content": "

@phunter Thank you very much again p! The question is - at my current physics level - which are these? :)

", "votes": null, "replies": [] }, { "id": 95875, "author_name": "phunter", "author_url": "", "post_date": "10/12/2015 19:18:31", "content": "

[quote=George Solymosi;95872]

\n\n

@phunter Thank you very much again p! The question is - at my current physics level - which are these? :)

\n\n

[/quote]\nfeatures with pt and p correlations may imply some mass correlation, like tau pt vs the momentum of the other 3 particles, I guess.

", "votes": null, "replies": [] }, { "id": 95880, "author_name": "", "author_url": "", "post_date": "10/12/2015 19:36:07", "content": "

Thank you I thought about pt and dira.

", "votes": null, "replies": [] }, { "id": 95883, "author_name": "phunter", "author_url": "", "post_date": "10/12/2015 20:08:46", "content": "

[quote=George Solymosi;95880]

\n\n

Thank you I thought about pt and dira.

\n\n

[/quote]\npt and dira is fine, no problem, but pt with dira vs p0/p2/p3 (pt0/pt1/pt2) can give some mass correlation. Let's think of energy conservation where a flying particle decays into 3 other flying particles where we know the mass of the 3 one (they are muons).

", "votes": null, "replies": [] }, { "id": 95894, "author_name": "", "author_url": "", "post_date": "10/12/2015 21:04:57", "content": "

Thank you again, the key achievements of this (far not flawless) competition, that I learnt a lot, my things started to work at last and "met" great guys and scientists like you, Sir.

", "votes": null, "replies": [] }, { "id": 95896, "author_name": "phunter", "author_url": "", "post_date": "10/12/2015 21:09:42", "content": "

[quote=George Solymosi;95894]

\n\n

Thank you again, the key achievements of this (far not flawless) competition, that I learnt a lot, my things started to work at last and "met" great guys and scientists like you, Sir.

\n\n

[/quote]\nYou are so welcomed, and your score beats my score

", "votes": null, "replies": [] }, { "id": 95897, "author_name": "", "author_url": "", "post_date": "10/12/2015 21:18:39", "content": "

As I mentioned before my score is just an experiment of mine with some obvious trick, thus at the end of all I am sure your score will much more "physically sound" and probably on LB will be higher as well. ;)

", "votes": null, "replies": [] }, { "id": 95902, "author_name": "", "author_url": "", "post_date": "10/12/2015 22:11:30", "content": "

[quote=phunter;95896]\nYou are so welcomed, and your score beats my score\n[/quote]

\n\n

UPDATE: I have a physically sound (meaning: no special trick) 0.990149 already! :) I wish I had more time.

", "votes": null, "replies": [] }, { "id": 95903, "author_name": "phunter", "author_url": "", "post_date": "10/12/2015 22:16:12", "content": "

[quote=George Solymosi;95902]

\n\n

[quote=phunter;95896]\nYou are so welcomed, and your score beats my score\n[/quote]

\n\n

UPDATE: I have a physically sound 0.990149 already! :) I wish I had more time.

\n\n

[/quote]\nWould be very nice to read your solution after the competition, and meaningful physics features are interesting.

\n\n

I learned in this competition that UGrad was nice, but I knew it too later and couldn't have time for coding a grid search interface for searching the parameters. So my solution was only xgboost + RF which may not look good finally: I just like xgboost so much. I will fix the pip installation for xgboost after this competition.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "95854": "Has somebody - who want to share - some magic hint to reduce CVM?", "95859": "[quote=George Solymosi;95854]\r\n\r\nHas somebody - who want to share - some magic hint to reduce CVM? \r\n\r\n[/quote]\r\nmy hint for never worrying about CvM test: try removing features correlated with tau invariant mass. CvM test exams if the model depends on tau mass or not, because the control channel is Ds decay which has a small mass difference. \r\n\r\nUsing the current information, you **might** reconstruct the tau invariant mass (or something correlated with this). I tested it, and it worked nice with +0.003 to +0.005 AUC at least, but these features may not have physical meanings which are useless for the real life, so I choice not to have these features.", "95872": "phunter Thank you very much again p! The question is - at my current physics level - which are these? :)", "95875": "[quote=George Solymosi;95872]\r\n\r\n@phunter Thank you very much again p! The question is - at my current physics level - which are these? :)\r\n\r\n[/quote]\r\nfeatures with pt and p correlations may imply some mass correlation, like tau pt vs the momentum of the other 3 particles, I guess.", "95880": "Thank you I thought about pt and dira.", "95883": "[quote=George Solymosi;95880]\r\n\r\nThank you I thought about pt and dira.\r\n\r\n[/quote]\r\npt and dira is fine, no problem, but pt with dira vs p0/p2/p3 (pt0/pt1/pt2) can give some mass correlation. Let's think of energy conservation where a flying particle decays into 3 other flying particles where we know the mass of the 3 one (they are muons).", "95894": "Thank you again, the key achievements of this (far not flawless) competition, that I learnt a lot, my things started to work at last and \"met\" great guys and scientists like you, Sir.", "95896": "[quote=George Solymosi;95894]\r\n\r\nThank you again, the key achievements of this (far not flawless) competition, that I learnt a lot, my things started to work at last and \"met\" great guys and scientists like you, Sir.\r\n\r\n[/quote]\r\nYou are so welcomed, and your score beats my score", "95897": "As I mentioned before my score is just an experiment of mine with some obvious trick, thus at the end of all I am sure your score will much more \"physically sound\" and probably on LB will be higher as well. ;)", "95902": "[quote=phunter;95896]\r\nYou are so welcomed, and your score beats my score\r\n[/quote]\r\n\r\n**UPDATE:** I have a physically sound (meaning: no special trick) **0.990149** already! :) I wish I had more time.", "95903": "[quote=George Solymosi;95902]\r\n\r\n[quote=phunter;95896]\r\nYou are so welcomed, and your score beats my score\r\n[/quote]\r\n\r\n**UPDATE:** I have a physically sound **0.990149** already! :) I wish I had more time.\r\n\r\n\r\n[/quote]\r\nWould be very nice to read your solution after the competition, and meaningful physics features are interesting. \r\n\r\nI learned in this competition that UGrad was nice, but I knew it too later and couldn't have time for coding a grid search interface for searching the parameters. So my solution was only xgboost + RF which may not look good finally: I just like xgboost so much. I will fix the pip installation for xgboost after this competition." }, "source": "meta" }