{
  "id": 17005,
  "title": "Challenge insights",
  "url": "/competitions/flavours-of-physics/discussion/17005",
  "author_name": "Anaderi",
  "post_date": "2015-10-14T21:07:46.173000",
  "votes": 7,
  "comment_count": 0,
  "views": 319,
  "content": "<p>Dear participants of the &#171;Flavors of Physics challenge!</p>\n\n<p>The competition is over and private leaderboard is published, where we see many  extraordinary results.</p>\n\n<p>This is very impressive, and most of them are results of really hard work but unfortunately not all of those are sensible from Physics side.</p>\n\n<p>While we are waiting for submissions code evaluation results, we want participants to share their ideas and lessons learned from this challenge, to help researches at CERN targeted to find rare decays.</p>\n\n<p>Here is tentative list of questions you are welcome to share your insights on:</p>\n\n<ul>\n<li>What are the features you used? Did you use any physical background for feature engineering?</li>\n<li>What ML tools performed best? </li>\n<li>Have you hit correlations/KS checks? (Do you hate them? :) Which approaches/tools drive to increase of these metrics? What methods have you used to fight correlation / disagreement (or how you cheated these metrics?). </li>\n<li>Have you tried to fool the checks? How?</li>\n<li>Have you tried to invent something new by involving agreement dataset into learning? Feature engineering with auto check? Adding normalization data to train? Adding loss term to penalize disagreement? </li>\n<li>Have you learned something new from this challenge?</li>\n</ul>\n\n<p>Please share your experience on any of the topics above (regardless of your position at leaderboard!).\nIt would be very interesting to meet/discuss those topics in person, so please consider <a href=\"https://www.kaggle.com/c/flavours-of-physics/forums/t/17003/nips-workshop-on-applying-machine-learning-to-experimental-physics\">applying for NIPS workshop</a>. </p>\n\n<p>Thanks a lot for all your time and efforts, we hope you enjoyed this puzzle ultimately.\nStay tuned - announcement on Physics Prize application will follow shortly!</p>\n\n<p>Organizers.</p>",
  "messages": [
    {
      "id": 96158,
      "postDate": "2015-10-14T21:07:46.173Z",
      "content": "<p>Dear participants of the &#171;Flavors of Physics challenge!</p>\n\n<p>The competition is over and private leaderboard is published, where we see many  extraordinary results.</p>\n\n<p>This is very impressive, and most of them are results of really hard work but unfortunately not all of those are sensible from Physics side.</p>\n\n<p>While we are waiting for submissions code evaluation results, we want participants to share their ideas and lessons learned from this challenge, to help researches at CERN targeted to find rare decays.</p>\n\n<p>Here is tentative list of questions you are welcome to share your insights on:</p>\n\n<ul>\n<li>What are the features you used? Did you use any physical background for feature engineering?</li>\n<li>What ML tools performed best? </li>\n<li>Have you hit correlations/KS checks? (Do you hate them? :) Which approaches/tools drive to increase of these metrics? What methods have you used to fight correlation / disagreement (or how you cheated these metrics?). </li>\n<li>Have you tried to fool the checks? How?</li>\n<li>Have you tried to invent something new by involving agreement dataset into learning? Feature engineering with auto check? Adding normalization data to train? Adding loss term to penalize disagreement? </li>\n<li>Have you learned something new from this challenge?</li>\n</ul>\n\n<p>Please share your experience on any of the topics above (regardless of your position at leaderboard!).\nIt would be very interesting to meet/discuss those topics in person, so please consider <a href=\"https://www.kaggle.com/c/flavours-of-physics/forums/t/17003/nips-workshop-on-applying-machine-learning-to-experimental-physics\">applying for NIPS workshop</a>. </p>\n\n<p>Thanks a lot for all your time and efforts, we hope you enjoyed this puzzle ultimately.\nStay tuned - announcement on Physics Prize application will follow shortly!</p>\n\n<p>Organizers.</p>",
      "rawMarkdown": "Dear participants of the «Flavors of Physics challenge!\r\n\r\nThe competition is over and private leaderboard is published, where we see many  extraordinary results.\r\n\r\nThis is very impressive, and most of them are results of really hard work but unfortunately not all of those are sensible from Physics side.\r\n\r\nWhile we are waiting for submissions code evaluation results, we want participants to share their ideas and lessons learned from this challenge, to help researches at CERN targeted to find rare decays.\r\n\r\nHere is tentative list of questions you are welcome to share your insights on:\r\n\r\n* What are the features you used? Did you use any physical background for feature engineering?\r\n* What ML tools performed best? \r\n* Have you hit correlations/KS checks? (Do you hate them? :) Which approaches/tools drive to increase of these metrics? What methods have you used to fight correlation / disagreement (or how you cheated these metrics?). \r\n* Have you tried to fool the checks? How?\r\n* Have you tried to invent something new by involving agreement dataset into learning? Feature engineering with auto check? Adding normalization data to train? Adding loss term to penalize disagreement? \r\n* Have you learned something new from this challenge?\r\n\r\nPlease share your experience on any of the topics above (regardless of your position at leaderboard!).\r\nIt would be very interesting to meet/discuss those topics in person, so please consider [applying for NIPS workshop][1]. \r\n\r\nThanks a lot for all your time and efforts, we hope you enjoyed this puzzle ultimately.\r\nStay tuned - announcement on Physics Prize application will follow shortly!\r\n\r\n\r\nOrganizers.\r\n\r\n\r\n  [1]: https://www.kaggle.com/c/flavours-of-physics/forums/t/17003/nips-workshop-on-applying-machine-learning-to-experimental-physics",
      "votes": 7
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "96158": "Dear participants of the «Flavors of Physics challenge!\r\n\r\nThe competition is over and private leaderboard is published, where we see many  extraordinary results.\r\n\r\nThis is very impressive, and most of them are results of really hard work but unfortunately not all of those are sensible from Physics side.\r\n\r\nWhile we are waiting for submissions code evaluation results, we want participants to share their ideas and lessons learned from this challenge, to help researches at CERN targeted to find rare decays.\r\n\r\nHere is tentative list of questions you are welcome to share your insights on:\r\n\r\n* What are the features you used? Did you use any physical background for feature engineering?\r\n* What ML tools performed best? \r\n* Have you hit correlations/KS checks? (Do you hate them? :) Which approaches/tools drive to increase of these metrics? What methods have you used to fight correlation / disagreement (or how you cheated these metrics?). \r\n* Have you tried to fool the checks? How?\r\n* Have you tried to invent something new by involving agreement dataset into learning? Feature engineering with auto check? Adding normalization data to train? Adding loss term to penalize disagreement? \r\n* Have you learned something new from this challenge?\r\n\r\nPlease share your experience on any of the topics above (regardless of your position at leaderboard!).\r\nIt would be very interesting to meet/discuss those topics in person, so please consider [applying for NIPS workshop][1]. \r\n\r\nThanks a lot for all your time and efforts, we hope you enjoyed this puzzle ultimately.\r\nStay tuned - announcement on Physics Prize application will follow shortly!\r\n\r\n\r\nOrganizers.\r\n\r\n\r\n  [1]: https://www.kaggle.com/c/flavours-of-physics/forums/t/17003/nips-workshop-on-applying-machine-learning-to-experimental-physics"
  }
}