{
  "id": 16843,
  "title": "Correlation philosophical question ;-)",
  "url": "/competitions/flavours-of-physics/discussion/16843",
  "author_name": "Michal Wojcik",
  "post_date": "2015-10-05T16:53:21.297000",
  "votes": 0,
  "comment_count": 0,
  "views": 287,
  "content": "<p>I started this competition a few days ago and I have the following question. Mass is strongly correlated with signal - this is easy to notice that majority of signal=1 is in the range (1740, 1810). So any good classifier which prediction will be correlated with signal (and this is the goal) must be correlated with mass. prediction=1 must be over-represented in the range (1740, 1810). Not because mass information was used, but because of the fact that signal itself is correlated with mass. So any avoiding of mass correlation must spoil classifier - ergo you can not improve classifier not increasing mass correlation. This doesn't make much sense for me. What do I miss?  </p>",
  "messages": [
    {
      "id": 95167,
      "postDate": "2015-10-05T16:53:21.297Z",
      "content": "<p>I started this competition a few days ago and I have the following question. Mass is strongly correlated with signal - this is easy to notice that majority of signal=1 is in the range (1740, 1810). So any good classifier which prediction will be correlated with signal (and this is the goal) must be correlated with mass. prediction=1 must be over-represented in the range (1740, 1810). Not because mass information was used, but because of the fact that signal itself is correlated with mass. So any avoiding of mass correlation must spoil classifier - ergo you can not improve classifier not increasing mass correlation. This doesn't make much sense for me. What do I miss?  </p>",
      "rawMarkdown": "I started this competition a few days ago and I have the following question. Mass is strongly correlated with signal - this is easy to notice that majority of signal=1 is in the range (1740, 1810). So any good classifier which prediction will be correlated with signal (and this is the goal) must be correlated with mass. prediction=1 must be over-represented in the range (1740, 1810). Not because mass information was used, but because of the fact that signal itself is correlated with mass. So any avoiding of mass correlation must spoil classifier - ergo you can not improve classifier not increasing mass correlation. This doesn't make much sense for me. What do I miss?  "
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "95167": "I started this competition a few days ago and I have the following question. Mass is strongly correlated with signal - this is easy to notice that majority of signal=1 is in the range (1740, 1810). So any good classifier which prediction will be correlated with signal (and this is the goal) must be correlated with mass. prediction=1 must be over-represented in the range (1740, 1810). Not because mass information was used, but because of the fact that signal itself is correlated with mass. So any avoiding of mass correlation must spoil classifier - ergo you can not improve classifier not increasing mass correlation. This doesn't make much sense for me. What do I miss?  "
  }
}