{
  "id": 506669,
  "title": "Choose good practice",
  "url": "/competitions/image-matching-challenge-2024/discussion/506669",
  "author_name": "",
  "post_date": "2024-05-22T19:11:02.015906700Z",
  "votes": 11,
  "comment_count": 1,
  "views": 0,
  "content": "<p>One thing that is very tempting in competitions like this is to overfit the public LB. Trying to tune our parameters for the best LB score rather than what actually makes the most sense. The current leaderboard is only calculated with 32% of the data, and we already know that these reconstructions to 3d images can be a little random. Scores are very tight so making insignificant changes on your CV is arguably not worth being extremely passionate about over a similar solution. I would expect things to shift a bit with this, likely just a bit more than last year. So be weary and double question your decisions! I know I sure am!</p>",
  "messages": [
    {
      "id": "2829769",
      "postDate": "05/22/2024 19:11:02",
      "content": "<p>One thing that is very tempting in competitions like this is to overfit the public LB. Trying to tune our parameters for the best LB score rather than what actually makes the most sense. The current leaderboard is only calculated with 32% of the data, and we already know that these reconstructions to 3d images can be a little random. Scores are very tight so making insignificant changes on your CV is arguably not worth being extremely passionate about over a similar solution. I would expect things to shift a bit with this, likely just a bit more than last year. So be weary and double question your decisions! I know I sure am!</p>",
      "rawMarkdown": "One thing that is very tempting in competitions like this is to overfit the public LB. Trying to tune our parameters for the best LB score rather than what actually makes the most sense. The current leaderboard is only calculated with 32% of the data, and we already know that these reconstructions to 3d images can be a little random. Scores are very tight so making insignificant changes on your CV is arguably not worth being extremely passionate about over a similar solution. I would expect things to shift a bit with this, likely just a bit more than last year. So be weary and double question your decisions! I know I sure am!",
      "votes": null
    },
    {
      "id": "2830221",
      "postDate": "05/23/2024 04:48:54",
      "content": "<p>you are right ,just change the parameters of open notebook to overfit the lb may useless</p>",
      "rawMarkdown": "you are right ,just change the parameters of open notebook to overfit the lb may useless",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2830221,
      "author_name": "cokkies",
      "author_url": "",
      "post_date": "05/23/2024 04:48:54",
      "content": "<p>you are right ,just change the parameters of open notebook to overfit the lb may useless</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2829769": "One thing that is very tempting in competitions like this is to overfit the public LB. Trying to tune our parameters for the best LB score rather than what actually makes the most sense. The current leaderboard is only calculated with 32% of the data, and we already know that these reconstructions to 3d images can be a little random. Scores are very tight so making insignificant changes on your CV is arguably not worth being extremely passionate about over a similar solution. I would expect things to shift a bit with this, likely just a bit more than last year. So be weary and double question your decisions! I know I sure am!",
    "2830221": "you are right ,just change the parameters of open notebook to overfit the lb may useless"
  },
  "source": "meta"
}