{
  "id": 475432,
  "title": "0.043 to 0.755 : Amazing jump of 1052 places to 2nd place ",
  "url": "/competitions/blood-vessel-segmentation/discussion/475432",
  "author_name": "",
  "post_date": "2024-02-08T11:49:52.284717700Z",
  "votes": 9,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Refer to SenNet+ HOA : Hacking the Human Vasculature in 3D competition.<br>\n<a href=\"https://www.kaggle.com/competitions/blood-vessel-segmentation/\" target=\"_blank\">https://www.kaggle.com/competitions/blood-vessel-segmentation/</a></p>\n<ul>\n<li>I was reviewing the private LB and could see a   participant who got a huge jump from 1055 rank to 2nd rank.</li>\n<li>The public LB score is mere 0.043 but private LB score if 0.755! 67% of data, the model performs poorly but 33% of data, it gains enormously.  The public LB is almost close to the sample submission!</li>\n</ul>\n<p>I think the participant should share their solution details and enlighten about their amazing recovery.</p>\n<ul>\n<li>This is also an opportunity to debate whether such a model is good one because on 67% of test data, it is not doing very badly( mere score of 0.043). In a real  scenario, one can expect the new data from the public and private LB regions with equal probability. </li>\n</ul>",
  "messages": [
    {
      "id": "2642763",
      "postDate": "02/08/2024 11:49:52",
      "content": "<p>Refer to SenNet+ HOA : Hacking the Human Vasculature in 3D competition.<br>\n<a href=\"https://www.kaggle.com/competitions/blood-vessel-segmentation/\" target=\"_blank\">https://www.kaggle.com/competitions/blood-vessel-segmentation/</a></p>\n<ul>\n<li>I was reviewing the private LB and could see a   participant who got a huge jump from 1055 rank to 2nd rank.</li>\n<li>The public LB score is mere 0.043 but private LB score if 0.755! 67% of data, the model performs poorly but 33% of data, it gains enormously.  The public LB is almost close to the sample submission!</li>\n</ul>\n<p>I think the participant should share their solution details and enlighten about their amazing recovery.</p>\n<ul>\n<li>This is also an opportunity to debate whether such a model is good one because on 67% of test data, it is not doing very badly( mere score of 0.043). In a real  scenario, one can expect the new data from the public and private LB regions with equal probability. </li>\n</ul>",
      "rawMarkdown": "Refer to SenNet+ HOA : Hacking the Human Vasculature in 3D competition.\nhttps://www.kaggle.com/competitions/blood-vessel-segmentation/\n- I was reviewing the private LB and could see a   participant who got a huge jump from 1055 rank to 2nd rank.\n- The public LB score is mere 0.043 but private LB score if 0.755! 67% of data, the model performs poorly but 33% of data, it gains enormously.  The public LB is almost close to the sample submission!\n\nI think the participant should share their solution details and enlighten about their amazing recovery.\n- This is also an opportunity to debate whether such a model is good one because on 67% of test data, it is not doing very badly( mere score of 0.043). In a real  scenario, one can expect the new data from the public and private LB regions with equal probability.",
      "votes": null
    },
    {
      "id": "2642946",
      "postDate": "02/08/2024 14:15:37",
      "content": "<p><a href=\"https://www.kaggle.com/crsuthikshnkumar\" target=\"_blank\">@crsuthikshnkumar</a> I agree with your point, this solution is worth reading for sure. Perhaps evaluating these models on yet another batch of unseen data will yield very different results.</p>",
      "rawMarkdown": "crsuthikshnkumar I agree with your point, this solution is worth reading for sure. Perhaps evaluating these models on yet another batch of unseen data will yield very different results.",
      "votes": null
    },
    {
      "id": "2644097",
      "postDate": "02/09/2024 09:33:15",
      "content": "<p>Please check <a href=\"https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/475657\" target=\"_blank\">https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/475657</a>.</p>\n<p>Maybe public/private data is so different. Jump up reason? it's just LUCK…</p>",
      "rawMarkdown": "Please check https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/475657.\n\nMaybe public/private data is so different. Jump up reason? it's just LUCK...",
      "votes": null
    },
    {
      "id": "2668975",
      "postDate": "02/26/2024 03:32:28",
      "content": "<p>metric/dataset relationship </p>",
      "rawMarkdown": "metric/dataset relationship",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2642946,
      "author_name": "ravi20076",
      "author_url": "",
      "post_date": "02/08/2024 14:15:37",
      "content": "<p><a href=\"https://www.kaggle.com/crsuthikshnkumar\" target=\"_blank\">@crsuthikshnkumar</a> I agree with your point, this solution is worth reading for sure. Perhaps evaluating these models on yet another batch of unseen data will yield very different results.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2644097,
          "author_name": "ojimaryoji",
          "author_url": "",
          "post_date": "02/09/2024 09:33:15",
          "content": "<p>Please check <a href=\"https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/475657\" target=\"_blank\">https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/475657</a>.</p>\n<p>Maybe public/private data is so different. Jump up reason? it's just LUCK…</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2668975,
      "author_name": "dragonzhang",
      "author_url": "",
      "post_date": "02/26/2024 03:32:28",
      "content": "<p>metric/dataset relationship </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2642763": "Refer to SenNet+ HOA : Hacking the Human Vasculature in 3D competition.\nhttps://www.kaggle.com/competitions/blood-vessel-segmentation/\n- I was reviewing the private LB and could see a   participant who got a huge jump from 1055 rank to 2nd rank.\n- The public LB score is mere 0.043 but private LB score if 0.755! 67% of data, the model performs poorly but 33% of data, it gains enormously.  The public LB is almost close to the sample submission!\n\nI think the participant should share their solution details and enlighten about their amazing recovery.\n- This is also an opportunity to debate whether such a model is good one because on 67% of test data, it is not doing very badly( mere score of 0.043). In a real  scenario, one can expect the new data from the public and private LB regions with equal probability.",
    "2642946": "crsuthikshnkumar I agree with your point, this solution is worth reading for sure. Perhaps evaluating these models on yet another batch of unseen data will yield very different results.",
    "2644097": "Please check https://www.kaggle.com/competitions/blood-vessel-segmentation/discussion/475657.\n\nMaybe public/private data is so different. Jump up reason? it's just LUCK...",
    "2668975": "metric/dataset relationship"
  },
  "source": "meta"
}