{
  "id": 37375,
  "title": "Episode 2. redux.",
  "url": "/competitions/dark-matter-signal-search-episode-2/discussion/37375",
  "author_name": "Anaderi",
  "post_date": "2017-08-01T07:21:54.785000",
  "votes": 0,
  "comment_count": 0,
  "views": 23,
  "content": "<p>Dear participants,</p>\n\n<p>after considering the winning solution we’ve realised that there was a flaw in the data. Background sample of the testing set had non-negligible amount of basetracks common with the training set. So the winning solution had exploited that fact by building a jungle of decision trees that have learned those background tracks. Unfortunately such solution doesn’t make any sense from physics point of view. So we’ve re-generated the samples, double checked it and extended the challenge a bit - until the end of the summer:</p>\n\n<p><a href=\"https://inclass.kaggle.com/c/dmse-2-1\">https://inclass.kaggle.com/c/dmse-2-1</a></p>\n\n<p>here is the link to the 3rd place solution by Konstantin Androsov that you can take advantage of:</p>\n\n<p><a href=\"https://github.com/kandrosov/mlhep2017/tree/work/episode2\">https://github.com/kandrosov/mlhep2017/tree/work/episode2</a></p>\n\n<p>I’m sorry for the mess, and hopefully this time you could have more fun and practice playing with the data. </p>",
  "messages": [
    {
      "id": 209098,
      "postDate": "2017-08-01T07:21:54.787Z",
      "content": "<p>Dear participants,</p>\n\n<p>after considering the winning solution we’ve realised that there was a flaw in the data. Background sample of the testing set had non-negligible amount of basetracks common with the training set. So the winning solution had exploited that fact by building a jungle of decision trees that have learned those background tracks. Unfortunately such solution doesn’t make any sense from physics point of view. So we’ve re-generated the samples, double checked it and extended the challenge a bit - until the end of the summer:</p>\n\n<p><a href=\"https://inclass.kaggle.com/c/dmse-2-1\">https://inclass.kaggle.com/c/dmse-2-1</a></p>\n\n<p>here is the link to the 3rd place solution by Konstantin Androsov that you can take advantage of:</p>\n\n<p><a href=\"https://github.com/kandrosov/mlhep2017/tree/work/episode2\">https://github.com/kandrosov/mlhep2017/tree/work/episode2</a></p>\n\n<p>I’m sorry for the mess, and hopefully this time you could have more fun and practice playing with the data. </p>",
      "rawMarkdown": "Dear participants,\r\n\r\nafter considering the winning solution we’ve realised that there was a flaw in the data. Background sample of the testing set had non-negligible amount of basetracks common with the training set. So the winning solution had exploited that fact by building a jungle of decision trees that have learned those background tracks. Unfortunately such solution doesn’t make any sense from physics point of view. So we’ve re-generated the samples, double checked it and extended the challenge a bit - until the end of the summer:\r\n\r\nhttps://inclass.kaggle.com/c/dmse-2-1\r\n\r\nhere is the link to the 3rd place solution by Konstantin Androsov that you can take advantage of:\r\n\r\nhttps://github.com/kandrosov/mlhep2017/tree/work/episode2\r\n\r\nI’m sorry for the mess, and hopefully this time you could have more fun and practice playing with the data. \r\n"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "209098": "Dear participants,\r\n\r\nafter considering the winning solution we’ve realised that there was a flaw in the data. Background sample of the testing set had non-negligible amount of basetracks common with the training set. So the winning solution had exploited that fact by building a jungle of decision trees that have learned those background tracks. Unfortunately such solution doesn’t make any sense from physics point of view. So we’ve re-generated the samples, double checked it and extended the challenge a bit - until the end of the summer:\r\n\r\nhttps://inclass.kaggle.com/c/dmse-2-1\r\n\r\nhere is the link to the 3rd place solution by Konstantin Androsov that you can take advantage of:\r\n\r\nhttps://github.com/kandrosov/mlhep2017/tree/work/episode2\r\n\r\nI’m sorry for the mess, and hopefully this time you could have more fun and practice playing with the data. \r\n"
  }
}