{
  "id": 390978,
  "title": "Just Curiosity !! Can we get best threshold values which doesn't overfit and worth it in private lb!!",
  "url": "/competitions/rsna-breast-cancer-detection/discussion/390978",
  "author_name": "Himanshu Nayal",
  "post_date": "2023-02-28T02:55:29.207000",
  "votes": 0,
  "comment_count": 7,
  "views": 0,
  "content": "<p>In starting of this competition there was a big discussion about the threshold value. Can we get the most extreme threshold value which didn't overfit on private data. </p>",
  "messages": [
    {
      "id": 2162333,
      "postDate": "2023-02-28T06:16:49.317Z",
      "content": "<p>I only submit 2 times and use threshold 0.5.</p>",
      "rawMarkdown": "I only submit 2 times and use threshold 0.5.",
      "votes": 1
    },
    {
      "id": 2162300,
      "postDate": "2023-02-28T05:39:01.617Z",
      "content": "<p>In my test, I always use 0.5, and its a little weird that no matter what best threshold value I got in CV, its almost always 0.5 to get best score in LB.</p>",
      "rawMarkdown": "In my test, I always use 0.5, and its a little weird that no matter what best threshold value I got in CV, its almost always 0.5 to get best score in LB.",
      "votes": 2
    },
    {
      "id": 2162197,
      "postDate": "2023-02-28T03:38:33.623Z",
      "content": "<p>We always apply the threshold for the best score on CV. Never prob the threshold on public LB, it will cause a huge shake.</p>",
      "rawMarkdown": "We always apply the threshold for the best score on CV. Never prob the threshold on public LB, it will cause a huge shake.",
      "votes": 2,
      "replies": [
        {
          "id": 2162233,
          "postDate": "2023-02-28T04:24:24.273Z",
          "content": "<p>True, I was curious about the threshold corresponding to the cv.Like how much thresholding helps or does it</p>",
          "rawMarkdown": "True, I was curious about the threshold corresponding to the cv.Like how much thresholding helps or does it",
          "replies": [
            {
              "id": 2162240,
              "postDate": "2023-02-28T04:28:52.870Z",
              "content": "<p>It depends on if your model is stable. We tried image-level models and multi-view models, the threshold of multi-view models are much more stable.</p>",
              "rawMarkdown": "It depends on if your model is stable. We tried image-level models and multi-view models, the threshold of multi-view models are much more stable."
            },
            {
              "id": 2162261,
              "postDate": "2023-02-28T04:43:31.237Z",
              "content": "<p>Agree with you and congratulations . See you in next competition.</p>",
              "rawMarkdown": "Agree with you and congratulations . See you in next competition.",
              "votes": 1
            },
            {
              "id": 2162483,
              "postDate": "2023-02-28T08:28:14.483Z",
              "content": "<p>Cheers cheers</p>",
              "rawMarkdown": "Cheers cheers"
            }
          ]
        }
      ]
    },
    {
      "id": 2162141,
      "postDate": "2023-02-28T02:55:29.207Z",
      "content": "<p>In starting of this competition there was a big discussion about the threshold value. Can we get the most extreme threshold value which didn't overfit on private data. </p>",
      "rawMarkdown": "In starting of this competition there was a big discussion about the threshold value. Can we get the most extreme threshold value which didn't overfit on private data. \n "
    }
  ],
  "comments": [
    {
      "id": 2162333,
      "author_name": "Anh Bui",
      "author_url": "",
      "post_date": "2023-02-28T06:16:49.317000",
      "content": "<p>I only submit 2 times and use threshold 0.5.</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 2162300,
      "author_name": "Mr.Fire",
      "author_url": "",
      "post_date": "2023-02-28T05:39:01.617000",
      "content": "<p>In my test, I always use 0.5, and its a little weird that no matter what best threshold value I got in CV, its almost always 0.5 to get best score in LB.</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 2162197,
      "author_name": "ForcewithMe",
      "author_url": "",
      "post_date": "2023-02-28T03:38:33.623000",
      "content": "<p>We always apply the threshold for the best score on CV. Never prob the threshold on public LB, it will cause a huge shake.</p>",
      "votes": 2,
      "replies": [
        {
          "id": 2162233,
          "author_name": "Himanshu Nayal",
          "author_url": "",
          "post_date": "2023-02-28T04:24:24.273000",
          "content": "<p>True, I was curious about the threshold corresponding to the cv.Like how much thresholding helps or does it</p>",
          "votes": 0,
          "replies": [
            {
              "id": 2162240,
              "author_name": "ForcewithMe",
              "author_url": "",
              "post_date": "2023-02-28T04:28:52.870000",
              "content": "<p>It depends on if your model is stable. We tried image-level models and multi-view models, the threshold of multi-view models are much more stable.</p>",
              "votes": 0,
              "replies": []
            },
            {
              "id": 2162261,
              "author_name": "Himanshu Nayal",
              "author_url": "",
              "post_date": "2023-02-28T04:43:31.237000",
              "content": "<p>Agree with you and congratulations . See you in next competition.</p>",
              "votes": 1,
              "replies": []
            },
            {
              "id": 2162483,
              "author_name": "ForcewithMe",
              "author_url": "",
              "post_date": "2023-02-28T08:28:14.483000",
              "content": "<p>Cheers cheers</p>",
              "votes": 0,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2162333": "I only submit 2 times and use threshold 0.5.",
    "2162300": "In my test, I always use 0.5, and its a little weird that no matter what best threshold value I got in CV, its almost always 0.5 to get best score in LB.",
    "2162197": "We always apply the threshold for the best score on CV. Never prob the threshold on public LB, it will cause a huge shake.",
    "2162141": "In starting of this competition there was a big discussion about the threshold value. Can we get the most extreme threshold value which didn't overfit on private data. \n "
  }
}