{
  "id": 306039,
  "title": "public score vs private score🔥🔥🔥",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/306039",
  "author_name": "",
  "post_date": "2022-02-08T00:25:41.742418200Z",
  "votes": 9,
  "comment_count": 8,
  "views": 0,
  "content": "<p>It is already proven that high resolution is high ranked at public score.</p>\n<p>But what choice should I make in the private score?</p>\n<p>What metric do you look at and decide which notebook to final submit?🤔</p>\n<p>Do you think there will be a big shake-up in this competition?</p>",
  "messages": [
    {
      "id": "1680655",
      "postDate": "02/08/2022 00:25:41",
      "content": "<p>It is already proven that high resolution is high ranked at public score.</p>\n<p>But what choice should I make in the private score?</p>\n<p>What metric do you look at and decide which notebook to final submit?🤔</p>\n<p>Do you think there will be a big shake-up in this competition?</p>",
      "rawMarkdown": "It is already proven that high resolution is high ranked at public score.\n\nBut what choice should I make in the private score?\n\nWhat metric do you look at and decide which notebook to final submit?🤔\n\nDo you think there will be a big shake-up in this competition?",
      "votes": null
    },
    {
      "id": "1680743",
      "postDate": "02/08/2022 02:20:38",
      "content": "<p>Of course.  It depends on model robustness.<br>\nThe jigsaw competition shakes up a lot.</p>\n<p>I think the final ranking simply   based on private score is fair enough.  To select models for submission creates problem.</p>\n<p>but the top ranking in this competition won't change much, in my opinion. </p>",
      "rawMarkdown": "Of course.  It depends on model robustness.\nThe jigsaw competition shakes up a lot.\n\nI think the final ranking simply   based on private score is fair enough.  To select models for submission creates problem.\n \nbut the top ranking in this competition won't change much, in my opinion.",
      "votes": null
    },
    {
      "id": "1680815",
      "postDate": "02/08/2022 03:27:34",
      "content": "<p>do not rely on the public LB. The public LB only indicates that your model is good at the extreamly small starfishes ( maybe 500 ~ 1500 ).<br>\nI don't expect the private 75% data is with the same distribution as the public 25%, cause none of the 3 videos of training set has that distribution, somehow i think the public 25% are chosen on purpose.</p>",
      "rawMarkdown": "do not rely on the public LB. The public LB only indicates that your model is good at the extreamly small starfishes ( maybe 500 ~ 1500 ).\nI don't expect the private 75% data is with the same distribution as the public 25%, cause none of the 3 videos of training set has that distribution, somehow i think the public 25% are chosen on purpose.",
      "votes": null
    },
    {
      "id": "1680882",
      "postDate": "02/08/2022 04:29:10",
      "content": "<p>Then, what matrix do you judge for the final submission?</p>",
      "rawMarkdown": "Then, what matrix do you judge for the final submission?",
      "votes": null
    },
    {
      "id": "1680887",
      "postDate": "02/08/2022 04:31:31",
      "content": "<p>Right, there was an amazing shake-up at The jigsaw competition.<br>\nmy leaderboard public score ranked 40th, but my private score disappeared into space.😂😂😂</p>",
      "rawMarkdown": "Right, there was an amazing shake-up at The jigsaw competition.\nmy leaderboard public score ranked 40th, but my private score disappeared into space.😂😂😂",
      "votes": null
    },
    {
      "id": "1680914",
      "postDate": "02/08/2022 04:59:32",
      "content": "<p>my best private score in Jigsaw in private bronze zone, while overfitted public \"FAKE\" score in public silver zone.<br>\nlol.<br>\nthe leading public scared me to choose overfitted submission, my best private score performs much worse in public score.</p>",
      "rawMarkdown": "my best private score in Jigsaw in private bronze zone, while overfitted public \"FAKE\" score in public silver zone.\nlol.\nthe leading public scared me to choose overfitted submission, my best private score performs much worse in public score.",
      "votes": null
    },
    {
      "id": "1680943",
      "postDate": "02/08/2022 05:25:59",
      "content": "<p>F2 of course, consider both CV and LB, but CV with more weight.</p>",
      "rawMarkdown": "F2 of course, consider both CV and LB, but CV with more weight.",
      "votes": null
    },
    {
      "id": "1680993",
      "postDate": "02/08/2022 06:37:53",
      "content": "<p>That's too bad😧 It is important to make a generalized model. But it's too hard..</p>",
      "rawMarkdown": "That's too bad😧 It is important to make a generalized model. But it's too hard..",
      "votes": null
    },
    {
      "id": "1680994",
      "postDate": "02/08/2022 06:39:10",
      "content": "<p>Thank you! I should check it out.</p>",
      "rawMarkdown": "Thank you! I should check it out.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1680743,
      "author_name": "dragonzhang",
      "author_url": "",
      "post_date": "02/08/2022 02:20:38",
      "content": "<p>Of course.  It depends on model robustness.<br>\nThe jigsaw competition shakes up a lot.</p>\n<p>I think the final ranking simply   based on private score is fair enough.  To select models for submission creates problem.</p>\n<p>but the top ranking in this competition won't change much, in my opinion. </p>",
      "votes": null,
      "replies": [
        {
          "id": 1680887,
          "author_name": "wonjunpark",
          "author_url": "",
          "post_date": "02/08/2022 04:31:31",
          "content": "<p>Right, there was an amazing shake-up at The jigsaw competition.<br>\nmy leaderboard public score ranked 40th, but my private score disappeared into space.😂😂😂</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1680914,
          "author_name": "dragonzhang",
          "author_url": "",
          "post_date": "02/08/2022 04:59:32",
          "content": "<p>my best private score in Jigsaw in private bronze zone, while overfitted public \"FAKE\" score in public silver zone.<br>\nlol.<br>\nthe leading public scared me to choose overfitted submission, my best private score performs much worse in public score.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1680993,
          "author_name": "wonjunpark",
          "author_url": "",
          "post_date": "02/08/2022 06:37:53",
          "content": "<p>That's too bad😧 It is important to make a generalized model. But it's too hard..</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1680815,
      "author_name": "snaker",
      "author_url": "",
      "post_date": "02/08/2022 03:27:34",
      "content": "<p>do not rely on the public LB. The public LB only indicates that your model is good at the extreamly small starfishes ( maybe 500 ~ 1500 ).<br>\nI don't expect the private 75% data is with the same distribution as the public 25%, cause none of the 3 videos of training set has that distribution, somehow i think the public 25% are chosen on purpose.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1680882,
          "author_name": "wonjunpark",
          "author_url": "",
          "post_date": "02/08/2022 04:29:10",
          "content": "<p>Then, what matrix do you judge for the final submission?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1680943,
          "author_name": "snaker",
          "author_url": "",
          "post_date": "02/08/2022 05:25:59",
          "content": "<p>F2 of course, consider both CV and LB, but CV with more weight.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1680994,
          "author_name": "wonjunpark",
          "author_url": "",
          "post_date": "02/08/2022 06:39:10",
          "content": "<p>Thank you! I should check it out.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1680655": "It is already proven that high resolution is high ranked at public score.\n\nBut what choice should I make in the private score?\n\nWhat metric do you look at and decide which notebook to final submit?🤔\n\nDo you think there will be a big shake-up in this competition?",
    "1680743": "Of course.  It depends on model robustness.\nThe jigsaw competition shakes up a lot.\n\nI think the final ranking simply   based on private score is fair enough.  To select models for submission creates problem.\n \nbut the top ranking in this competition won't change much, in my opinion.",
    "1680815": "do not rely on the public LB. The public LB only indicates that your model is good at the extreamly small starfishes ( maybe 500 ~ 1500 ).\nI don't expect the private 75% data is with the same distribution as the public 25%, cause none of the 3 videos of training set has that distribution, somehow i think the public 25% are chosen on purpose.",
    "1680882": "Then, what matrix do you judge for the final submission?",
    "1680887": "Right, there was an amazing shake-up at The jigsaw competition.\nmy leaderboard public score ranked 40th, but my private score disappeared into space.😂😂😂",
    "1680914": "my best private score in Jigsaw in private bronze zone, while overfitted public \"FAKE\" score in public silver zone.\nlol.\nthe leading public scared me to choose overfitted submission, my best private score performs much worse in public score.",
    "1680943": "F2 of course, consider both CV and LB, but CV with more weight.",
    "1680993": "That's too bad😧 It is important to make a generalized model. But it's too hard..",
    "1680994": "Thank you! I should check it out."
  },
  "source": "meta"
}