{
  "id": 307480,
  "title": "Will it Shake up like \"Jigsaw Rate Severity of Toxic Comments\"?",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/307480",
  "author_name": "fxalll",
  "post_date": "2022-02-14T10:59:45.731000",
  "votes": 7,
  "comment_count": 6,
  "views": 0,
  "content": "<p>In \"Jigsaw Rate Severity of Toxic Comments\", the private leaderboard is calculated with approximately 95% of the test data . I even saw the 20th place fall to the 2000th 😨.<br>\nIn this Competition, the final results will be based on the other 75%. I wonder if there is over-fitting in the high score public LB 👀</p>",
  "messages": [
    {
      "id": 1689600,
      "postDate": "2022-02-14T10:59:45.730Z",
      "content": "<p>In \"Jigsaw Rate Severity of Toxic Comments\", the private leaderboard is calculated with approximately 95% of the test data . I even saw the 20th place fall to the 2000th 😨.<br>\nIn this Competition, the final results will be based on the other 75%. I wonder if there is over-fitting in the high score public LB 👀</p>",
      "rawMarkdown": "In \"Jigsaw Rate Severity of Toxic Comments\", the private leaderboard is calculated with approximately 95% of the test data . I even saw the 20th place fall to the 2000th 😨.\nIn this Competition, the final results will be based on the other 75%. I wonder if there is over-fitting in the high score public LB 👀",
      "votes": 7
    },
    {
      "id": 1689617,
      "postDate": "2022-02-14T11:16:50.060Z",
      "content": "<p><a href=\"https://www.kaggle.com/Fxalll\" target=\"_blank\">@Fxalll</a>: There was a shakeup in Jigsaw for the simple reason that the public LB score was multiplied by a factor. Whoever multiplied with a factor, score improved in public LB and pushed down in private LB. Whoever didn't and trusted their CV, the score was consistent with private LB.</p>\n<p>Even for this competition, if public LB score and CV score correlates, I don't think there would be a huge shakeup.  </p>",
      "rawMarkdown": "@Fxalll: There was a shakeup in Jigsaw for the simple reason that the public LB score was multiplied by a factor. Whoever multiplied with a factor, score improved in public LB and pushed down in private LB. Whoever didn't and trusted their CV, the score was consistent with private LB.\n\nEven for this competition, if public LB score and CV score correlates, I don't think there would be a huge shakeup.  ",
      "votes": 2,
      "replies": [
        {
          "id": 1689648,
          "postDate": "2022-02-14T11:40:36.990Z",
          "content": "<p>You are right, I also found those factors confusing. And I think LB may have used deliberately selected data, which will cause the ranking to shake</p>",
          "rawMarkdown": "You are right, I also found those factors confusing. And I think LB may have used deliberately selected data, which will cause the ranking to shake",
          "votes": 1
        },
        {
          "id": 1690500,
          "postDate": "2022-02-15T01:02:23.687Z",
          "content": "<p>Yup, we've now been shaken up…</p>",
          "rawMarkdown": "Yup, we've now been shaken up..."
        }
      ]
    },
    {
      "id": 1689686,
      "postDate": "2022-02-14T12:00:41.327Z",
      "content": "<p>I don't think so. We are far from the first place's over-fitting result.🙈🙉🙊</p>",
      "rawMarkdown": "I don't think so. We are far from the first place's over-fitting result.🙈🙉🙊",
      "replies": [
        {
          "id": 1689691,
          "postDate": "2022-02-14T12:08:50.597Z",
          "content": "<p>LB &gt; 0.8 is amazing</p>",
          "rawMarkdown": "LB > 0.8 is amazing"
        }
      ]
    },
    {
      "id": 1690495,
      "postDate": "2022-02-15T00:59:20.737Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1689617,
      "author_name": "Manoj Prabhakar",
      "author_url": "",
      "post_date": "2022-02-14T11:16:50.060000",
      "content": "<p><a href=\"https://www.kaggle.com/Fxalll\" target=\"_blank\">@Fxalll</a>: There was a shakeup in Jigsaw for the simple reason that the public LB score was multiplied by a factor. Whoever multiplied with a factor, score improved in public LB and pushed down in private LB. Whoever didn't and trusted their CV, the score was consistent with private LB.</p>\n<p>Even for this competition, if public LB score and CV score correlates, I don't think there would be a huge shakeup.  </p>",
      "votes": 2,
      "replies": [
        {
          "id": 1689648,
          "author_name": "fxalll",
          "author_url": "",
          "post_date": "2022-02-14T11:40:36.990000",
          "content": "<p>You are right, I also found those factors confusing. And I think LB may have used deliberately selected data, which will cause the ranking to shake</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 1690500,
          "author_name": "Dino Wun",
          "author_url": "",
          "post_date": "2022-02-15T01:02:23.687000",
          "content": "<p>Yup, we've now been shaken up…</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1689686,
      "author_name": "Phoenix Ash",
      "author_url": "",
      "post_date": "2022-02-14T12:00:41.327000",
      "content": "<p>I don't think so. We are far from the first place's over-fitting result.🙈🙉🙊</p>",
      "votes": 0,
      "replies": [
        {
          "id": 1689691,
          "author_name": "fxalll",
          "author_url": "",
          "post_date": "2022-02-14T12:08:50.597000",
          "content": "<p>LB &gt; 0.8 is amazing</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1690495,
      "author_name": "",
      "author_url": "",
      "post_date": "2022-02-15T00:59:20.737000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1689600": "In \"Jigsaw Rate Severity of Toxic Comments\", the private leaderboard is calculated with approximately 95% of the test data . I even saw the 20th place fall to the 2000th 😨.\nIn this Competition, the final results will be based on the other 75%. I wonder if there is over-fitting in the high score public LB 👀",
    "1689617": "@Fxalll: There was a shakeup in Jigsaw for the simple reason that the public LB score was multiplied by a factor. Whoever multiplied with a factor, score improved in public LB and pushed down in private LB. Whoever didn't and trusted their CV, the score was consistent with private LB.\n\nEven for this competition, if public LB score and CV score correlates, I don't think there would be a huge shakeup.  ",
    "1689686": "I don't think so. We are far from the first place's over-fitting result.🙈🙉🙊",
    "1690495": ""
  }
}