{
  "id": 12496,
  "title": "Value of a top solution post contest",
  "url": "/competitions/avazu-ctr-prediction/discussion/12496",
  "author_name": "",
  "post_date": "2015-02-12T18:37:46.230Z",
  "votes": null,
  "comment_count": 1,
  "views": 1069,
  "content": "<p>I'm sure this happens from time to time, someone working on a now defunct contest finds a way to get a substantially better result. Is there any outlet for said solution (beyond publish in forums that are probably not read much a week after said contest is ended)? Does the company sponsoring it want/have any desire for such a thing or is it all academic?</p>",
  "messages": [
    {
      "id": "64137",
      "postDate": "02/12/2015 18:37:46",
      "content": "<p>I'm sure this happens from time to time, someone working on a now defunct contest finds a way to get a substantially better result. Is there any outlet for said solution (beyond publish in forums that are probably not read much a week after said contest is ended)? Does the company sponsoring it want/have any desire for such a thing or is it all academic?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "64327",
      "postDate": "02/16/2015 08:40:02",
      "content": "<p>I have 2 related questions:</p>\n<p>1) How can i see the post-competition private/public leaderboard for the submissions sent after the deadline in a given kaggle competition?<br><br> 2) In this specific competition would some company be interested in a say 0.347 solution?. <br>If you count the bits needed to go from the 0.45 trivial &quot;benchmark&quot; solution down to 0.379 winners solution and then count the supplementary bits needed to go to 0.347 you will know that this is not a rehash/optimization of current ideas floating in the field, there is a too big difference and also you can't expect to go down to near 0 in this field.<br> Also data is quite &quot;doctored&quot; in this competition's day 31 comparing with previous days ((on top of the already known holiday) in the following sense: probability of (assuming same random sampling) of day 31 happening is below 2^-5000 while between other consecutive days is above 0.01 in a reasonable optimal model) so one expects better numbers in production &quot;un-doctored&quot; data between train/test&nbsp; periods.<br>Or maybe the organizers should say at end of competition something like &quot;our best algorithm already gets 0.xxx on this combination of train/test data&quot;.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 64327,
      "author_name": "cubbus",
      "author_url": "",
      "post_date": "02/16/2015 08:40:02",
      "content": "<p>I have 2 related questions:</p>\n<p>1) How can i see the post-competition private/public leaderboard for the submissions sent after the deadline in a given kaggle competition?<br><br> 2) In this specific competition would some company be interested in a say 0.347 solution?. <br>If you count the bits needed to go from the 0.45 trivial &quot;benchmark&quot; solution down to 0.379 winners solution and then count the supplementary bits needed to go to 0.347 you will know that this is not a rehash/optimization of current ideas floating in the field, there is a too big difference and also you can't expect to go down to near 0 in this field.<br> Also data is quite &quot;doctored&quot; in this competition's day 31 comparing with previous days ((on top of the already known holiday) in the following sense: probability of (assuming same random sampling) of day 31 happening is below 2^-5000 while between other consecutive days is above 0.01 in a reasonable optimal model) so one expects better numbers in production &quot;un-doctored&quot; data between train/test&nbsp; periods.<br>Or maybe the organizers should say at end of competition something like &quot;our best algorithm already gets 0.xxx on this combination of train/test data&quot;.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "64137": "",
    "64327": ""
  },
  "source": "meta"
}