{
  "id": 287385,
  "title": "hosts should remove the overfitted PB",
  "url": "/competitions/g-research-crypto-forecasting/discussion/287385",
  "author_name": "dragon zhang",
  "post_date": "2021-11-13T15:33:14.780000",
  "votes": 8,
  "comment_count": 7,
  "views": 0,
  "content": "<p>Or at specific daily deadline , using real time data not included in train.csv to test and publish the score.</p>",
  "messages": [
    {
      "id": 1581286,
      "postDate": "2021-11-13T15:33:14.780Z",
      "content": "<p>Or at specific daily deadline , using real time data not included in train.csv to test and publish the score.</p>",
      "rawMarkdown": "Or at specific daily deadline , using real time data not included in train.csv to test and publish the score.",
      "votes": 8
    },
    {
      "id": 1581431,
      "postDate": "2021-11-13T18:54:00.213Z",
      "content": "<p>That's a very good idea and would solve a lot of problems but I think needs a lot of time and effort to rerun every kernel… maybe weekly updates are more doable but I doubt if they change that now. </p>\n<p>ps: anyway, at the end it's up to us how we use the LB, if you want to have an idea of your true models performance on unseen data just restrict yourself to train/submit with the overlapping samples - still we are missing the true LB ranking this way, and we don't know what the best possible (true) score has been achieved </p>",
      "rawMarkdown": "That's a very good idea and would solve a lot of problems but I think needs a lot of time and effort to rerun every kernel... maybe weekly updates are more doable but I doubt if they change that now. \n\nps: anyway, at the end it's up to us how we use the LB, if you want to have an idea of your true models performance on unseen data just restrict yourself to train/submit with the overlapping samples - still we are missing the true LB ranking this way, and we don't know what the best possible (true) score has been achieved ",
      "votes": 1,
      "replies": [
        {
          "id": 1581577,
          "postDate": "2021-11-14T01:04:50.027Z",
          "rawMarkdown": "",
          "isDeleted": true
        },
        {
          "id": 1581579,
          "postDate": "2021-11-14T01:07:57.707Z",
          "content": "<p>It happens often I have seen in other competitions myself</p>",
          "rawMarkdown": "It happens often I have seen in other competitions myself"
        }
      ]
    },
    {
      "id": 1622838,
      "postDate": "2021-12-19T07:38:54.860Z",
      "content": "<p>I am still confused on how more than 100 people have perfect scores in the LB. Can anyone give me context ?</p>",
      "rawMarkdown": "I am still confused on how more than 100 people have perfect scores in the LB. Can anyone give me context ?"
    },
    {
      "id": 1582037,
      "postDate": "2021-11-14T13:29:53.187Z",
      "content": "<p>Hello everyone</p>",
      "rawMarkdown": "Hello everyone"
    },
    {
      "id": 1581580,
      "postDate": "2021-11-14T01:11:14.973Z",
      "rawMarkdown": "",
      "isDeleted": true
    },
    {
      "id": 1581469,
      "postDate": "2021-11-13T20:21:49.640Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1581431,
      "author_name": "Ioannis M",
      "author_url": "",
      "post_date": "2021-11-13T18:54:00.213000",
      "content": "<p>That's a very good idea and would solve a lot of problems but I think needs a lot of time and effort to rerun every kernel… maybe weekly updates are more doable but I doubt if they change that now. </p>\n<p>ps: anyway, at the end it's up to us how we use the LB, if you want to have an idea of your true models performance on unseen data just restrict yourself to train/submit with the overlapping samples - still we are missing the true LB ranking this way, and we don't know what the best possible (true) score has been achieved </p>",
      "votes": 1,
      "replies": [
        {
          "id": 1581577,
          "author_name": "",
          "author_url": "",
          "post_date": "2021-11-14T01:04:50.027000",
          "content": "",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1581579,
          "author_name": "Sayantan Mazumdar",
          "author_url": "",
          "post_date": "2021-11-14T01:07:57.707000",
          "content": "<p>It happens often I have seen in other competitions myself</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1622838,
      "author_name": "Aditya Anil",
      "author_url": "",
      "post_date": "2021-12-19T07:38:54.860000",
      "content": "<p>I am still confused on how more than 100 people have perfect scores in the LB. Can anyone give me context ?</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1582037,
      "author_name": "sankha1234",
      "author_url": "",
      "post_date": "2021-11-14T13:29:53.187000",
      "content": "<p>Hello everyone</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1581580,
      "author_name": "",
      "author_url": "",
      "post_date": "2021-11-14T01:11:14.973000",
      "content": "",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1581469,
      "author_name": "",
      "author_url": "",
      "post_date": "2021-11-13T20:21:49.640000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1581286": "Or at specific daily deadline , using real time data not included in train.csv to test and publish the score.",
    "1581431": "That's a very good idea and would solve a lot of problems but I think needs a lot of time and effort to rerun every kernel... maybe weekly updates are more doable but I doubt if they change that now. \n\nps: anyway, at the end it's up to us how we use the LB, if you want to have an idea of your true models performance on unseen data just restrict yourself to train/submit with the overlapping samples - still we are missing the true LB ranking this way, and we don't know what the best possible (true) score has been achieved ",
    "1622838": "I am still confused on how more than 100 people have perfect scores in the LB. Can anyone give me context ?",
    "1582037": "Hello everyone",
    "1581580": "",
    "1581469": ""
  }
}