{
  "id": 411360,
  "title": "what percentage of data is in public and private test?",
  "url": "/competitions/vesuvius-challenge-ink-detection/discussion/411360",
  "author_name": "",
  "post_date": "2023-05-18T20:18:04.852010400Z",
  "votes": 2,
  "comment_count": 7,
  "views": 0,
  "content": "<p>\"This leaderboard is calculated with only a (smaller) portion of the test data. The final results will be based on the remaining (larger) portion, so the final standings may be different.\"</p>\n<p>unlike other previous compeition, the percentage of data in public and private test dat is npt disclosed.<br>\nBased on my esperimental results, it seems that the number of ink pixels in public set is very small.</p>\n<p>that is probably the reason for big jump in LB.<br>\nanyone has further insights?</p>",
  "messages": [
    {
      "id": "2264894",
      "postDate": "05/18/2023 20:18:04",
      "content": "<p>\"This leaderboard is calculated with only a (smaller) portion of the test data. The final results will be based on the remaining (larger) portion, so the final standings may be different.\"</p>\n<p>unlike other previous compeition, the percentage of data in public and private test dat is npt disclosed.<br>\nBased on my esperimental results, it seems that the number of ink pixels in public set is very small.</p>\n<p>that is probably the reason for big jump in LB.<br>\nanyone has further insights?</p>",
      "rawMarkdown": "\"This leaderboard is calculated with only a (smaller) portion of the test data. The final results will be based on the remaining (larger) portion, so the final standings may be different.\"\n\nunlike other previous compeition, the percentage of data in public and private test dat is npt disclosed.\nBased on my esperimental results, it seems that the number of ink pixels in public set is very small.\n\nthat is probably the reason for big jump in LB.\nanyone has further insights?",
      "votes": null
    },
    {
      "id": "2264900",
      "postDate": "05/18/2023 20:26:42",
      "content": "<p><a href=\"https://www.kaggle.com/competitions/vesuvius-challenge-ink-detection/discussion/401514\" target=\"_blank\">here </a></p>",
      "rawMarkdown": "[here ](https://www.kaggle.com/competitions/vesuvius-challenge-ink-detection/discussion/401514)",
      "votes": null
    },
    {
      "id": "2265000",
      "postDate": "05/18/2023 23:04:33",
      "content": "<p>thanks for the link</p>",
      "rawMarkdown": "thanks for the link",
      "votes": null
    },
    {
      "id": "2265096",
      "postDate": "05/19/2023 01:25:21",
      "content": "<p>The competition hosts said that the number of pixels is roughly 10%.  But, I think the number of pixels isn't necessarily the most important aspect.  The number of ink pixels seems to be more important.  If there are a lot of no ink pixels and only a few ink pixels, that would contribute a lot to an unreliable lb.  Were you able to get an approximate ratio of the number of predicted positive pixels to negative pixels?</p>",
      "rawMarkdown": "The competition hosts said that the number of pixels is roughly 10%.  But, I think the number of pixels isn't necessarily the most important aspect.  The number of ink pixels seems to be more important.  If there are a lot of no ink pixels and only a few ink pixels, that would contribute a lot to an unreliable lb.  Were you able to get an approximate ratio of the number of predicted positive pixels to negative pixels?",
      "votes": null
    },
    {
      "id": "2265130",
      "postDate": "05/19/2023 02:27:43",
      "content": "<p>\"Were you able to get an approximate ratio of the number of predicted positive pixels to negative pixels?\"</p>\n<p>becuase of  the of the dataset paper, ….<br>\na definitely yes!!!</p>",
      "rawMarkdown": "\"Were you able to get an approximate ratio of the number of predicted positive pixels to negative pixels?\"\n\nbecuase of  the of the dataset paper, ....\na definitely yes!!!",
      "votes": null
    },
    {
      "id": "2265881",
      "postDate": "05/19/2023 15:21:16",
      "content": "<p>We know that submitting the mask gives lb 0.11, so this is the best estimate of how much ink is there I guess. And it is fairly close to the percentage of ink area for the 3rd training fragment (which is a part of the same scroll).</p>",
      "rawMarkdown": "We know that submitting the mask gives lb 0.11, so this is the best estimate of how much ink is there I guess. And it is fairly close to the percentage of ink area for the 3rd training fragment (which is a part of the same scroll).",
      "votes": null
    },
    {
      "id": "2278045",
      "postDate": "05/28/2023 09:27:54",
      "content": "<p>Does that mean it is about 11%? </p>",
      "rawMarkdown": "Does that mean it is about 11%?",
      "votes": null
    },
    {
      "id": "2278974",
      "postDate": "05/29/2023 06:01:57",
      "content": "<p>yes, you can do calculation to calculate the ratio of tp and fp given recall = 1</p>",
      "rawMarkdown": "yes, you can do calculation to calculate the ratio of tp and fp given recall = 1",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2264900,
      "author_name": "iraqbot",
      "author_url": "",
      "post_date": "05/18/2023 20:26:42",
      "content": "<p><a href=\"https://www.kaggle.com/competitions/vesuvius-challenge-ink-detection/discussion/401514\" target=\"_blank\">here </a></p>",
      "votes": null,
      "replies": [
        {
          "id": 2265000,
          "author_name": "hengck23",
          "author_url": "",
          "post_date": "05/18/2023 23:04:33",
          "content": "<p>thanks for the link</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2265096,
      "author_name": "petersk20",
      "author_url": "",
      "post_date": "05/19/2023 01:25:21",
      "content": "<p>The competition hosts said that the number of pixels is roughly 10%.  But, I think the number of pixels isn't necessarily the most important aspect.  The number of ink pixels seems to be more important.  If there are a lot of no ink pixels and only a few ink pixels, that would contribute a lot to an unreliable lb.  Were you able to get an approximate ratio of the number of predicted positive pixels to negative pixels?</p>",
      "votes": null,
      "replies": [
        {
          "id": 2265130,
          "author_name": "hengck23",
          "author_url": "",
          "post_date": "05/19/2023 02:27:43",
          "content": "<p>\"Were you able to get an approximate ratio of the number of predicted positive pixels to negative pixels?\"</p>\n<p>becuase of  the of the dataset paper, ….<br>\na definitely yes!!!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2265881,
      "author_name": "alexz0",
      "author_url": "",
      "post_date": "05/19/2023 15:21:16",
      "content": "<p>We know that submitting the mask gives lb 0.11, so this is the best estimate of how much ink is there I guess. And it is fairly close to the percentage of ink area for the 3rd training fragment (which is a part of the same scroll).</p>",
      "votes": null,
      "replies": [
        {
          "id": 2278045,
          "author_name": "yassinealouini",
          "author_url": "",
          "post_date": "05/28/2023 09:27:54",
          "content": "<p>Does that mean it is about 11%? </p>",
          "votes": null,
          "replies": [
            {
              "id": 2278974,
              "author_name": "fengqilong",
              "author_url": "",
              "post_date": "05/29/2023 06:01:57",
              "content": "<p>yes, you can do calculation to calculate the ratio of tp and fp given recall = 1</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2264894": "\"This leaderboard is calculated with only a (smaller) portion of the test data. The final results will be based on the remaining (larger) portion, so the final standings may be different.\"\n\nunlike other previous compeition, the percentage of data in public and private test dat is npt disclosed.\nBased on my esperimental results, it seems that the number of ink pixels in public set is very small.\n\nthat is probably the reason for big jump in LB.\nanyone has further insights?",
    "2264900": "[here ](https://www.kaggle.com/competitions/vesuvius-challenge-ink-detection/discussion/401514)",
    "2265000": "thanks for the link",
    "2265096": "The competition hosts said that the number of pixels is roughly 10%.  But, I think the number of pixels isn't necessarily the most important aspect.  The number of ink pixels seems to be more important.  If there are a lot of no ink pixels and only a few ink pixels, that would contribute a lot to an unreliable lb.  Were you able to get an approximate ratio of the number of predicted positive pixels to negative pixels?",
    "2265130": "\"Were you able to get an approximate ratio of the number of predicted positive pixels to negative pixels?\"\n\nbecuase of  the of the dataset paper, ....\na definitely yes!!!",
    "2265881": "We know that submitting the mask gives lb 0.11, so this is the best estimate of how much ink is there I guess. And it is fairly close to the percentage of ink area for the 3rd training fragment (which is a part of the same scroll).",
    "2278045": "Does that mean it is about 11%?",
    "2278974": "yes, you can do calculation to calculate the ratio of tp and fp given recall = 1"
  },
  "source": "meta"
}