{
  "id": 225278,
  "title": "Wrong submission but higher score",
  "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/225278",
  "author_name": "",
  "post_date": "2021-03-11T15:18:00.754162600Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I have trained a model specifically for CVC, but I made some mistakes during the submission process so that my CVC model lost 0.08 weight. My CVC confidence value did not exceed 0.92, but the score was higher than normal submission. I think there is some kind of post-processing that can improve performance, but I have no experience. Would someone please provide some ideas?</p>",
  "messages": [
    {
      "id": "1234772",
      "postDate": "03/11/2021 15:18:00",
      "content": "<p>I have trained a model specifically for CVC, but I made some mistakes during the submission process so that my CVC model lost 0.08 weight. My CVC confidence value did not exceed 0.92, but the score was higher than normal submission. I think there is some kind of post-processing that can improve performance, but I have no experience. Would someone please provide some ideas?</p>",
      "rawMarkdown": "I have trained a model specifically for CVC, but I made some mistakes during the submission process so that my CVC model lost 0.08 weight. My CVC confidence value did not exceed 0.92, but the score was higher than normal submission. I think there is some kind of post-processing that can improve performance, but I have no experience. Would someone please provide some ideas?",
      "votes": null
    },
    {
      "id": "1234861",
      "postDate": "03/11/2021 16:18:33",
      "content": "<p>Something along this line ?  <a href=\"https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/220389\" target=\"_blank\">https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/220389</a><br>\nNot sure if it will work here though</p>",
      "rawMarkdown": "Something along this line ?  https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/220389\nNot sure if it will work here though",
      "votes": null
    },
    {
      "id": "1235284",
      "postDate": "03/12/2021 03:08:26",
      "content": "<p>It is unlikely that really made much impact because auc does not care about the overall magnitude, just the sort order. If you are ensembling multiple models then yes it can have an impact as the various models interact, but simply shifting or rescaling the distribution cannot actually change performance. </p>",
      "rawMarkdown": "It is unlikely that really made much impact because auc does not care about the overall magnitude, just the sort order. If you are ensembling multiple models then yes it can have an impact as the various models interact, but simply shifting or rescaling the distribution cannot actually change performance.",
      "votes": null
    },
    {
      "id": "1235826",
      "postDate": "03/12/2021 13:59:48",
      "content": "<p>Same thought… Auc seems is not sensitive to whole change to predict.</p>",
      "rawMarkdown": "Same thought... Auc seems is not sensitive to whole change to predict.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1234861,
      "author_name": "nyleve",
      "author_url": "",
      "post_date": "03/11/2021 16:18:33",
      "content": "<p>Something along this line ?  <a href=\"https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/220389\" target=\"_blank\">https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/220389</a><br>\nNot sure if it will work here though</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1235284,
      "author_name": "ryches",
      "author_url": "",
      "post_date": "03/12/2021 03:08:26",
      "content": "<p>It is unlikely that really made much impact because auc does not care about the overall magnitude, just the sort order. If you are ensembling multiple models then yes it can have an impact as the various models interact, but simply shifting or rescaling the distribution cannot actually change performance. </p>",
      "votes": null,
      "replies": [
        {
          "id": 1235826,
          "author_name": "fanwenping",
          "author_url": "",
          "post_date": "03/12/2021 13:59:48",
          "content": "<p>Same thought… Auc seems is not sensitive to whole change to predict.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1234772": "I have trained a model specifically for CVC, but I made some mistakes during the submission process so that my CVC model lost 0.08 weight. My CVC confidence value did not exceed 0.92, but the score was higher than normal submission. I think there is some kind of post-processing that can improve performance, but I have no experience. Would someone please provide some ideas?",
    "1234861": "Something along this line ?  https://www.kaggle.com/c/rfcx-species-audio-detection/discussion/220389\nNot sure if it will work here though",
    "1235284": "It is unlikely that really made much impact because auc does not care about the overall magnitude, just the sort order. If you are ensembling multiple models then yes it can have an impact as the various models interact, but simply shifting or rescaling the distribution cannot actually change performance.",
    "1235826": "Same thought... Auc seems is not sensitive to whole change to predict."
  },
  "source": "meta"
}