{
  "id": 459396,
  "title": "When predicting Step as correct if it lies in 30 min range for onset or wakeup, the algorithm got 89% accuracy. But when using AP scores, it got only 0.37%",
  "url": "/competitions/child-mind-institute-detect-sleep-states/discussion/459396",
  "author_name": "",
  "post_date": "2023-12-05T02:23:22.063892Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I was analysing possible obstacles, and found out that sometimes person can be active before Wakeup is reported (applied some filters to understand the problem), in other words, the exact Onset or Wakep Events  can be not so precise… Can you explain how these events were assigned?</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F17547704%2F6a6adb6addf633688d1561a876aa3d3c%2Fexample.png?generation=1701742942562588&amp;alt=media\" alt=\"\"></p>\n<p>And I also don't know how to generate confedence score. I took 1.0 everywhere. Can it make mistakes more drammatic in terms of scores?</p>\n<p>PS: 89% was scored on mostly unseen training data <br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F17547704%2F22f477e0482176c41c5d60c5b2efe1de%2Fexample2.png?generation=1701746152816745&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": "2549088",
      "postDate": "12/05/2023 02:23:22",
      "content": "<p>I was analysing possible obstacles, and found out that sometimes person can be active before Wakeup is reported (applied some filters to understand the problem), in other words, the exact Onset or Wakep Events  can be not so precise… Can you explain how these events were assigned?</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F17547704%2F6a6adb6addf633688d1561a876aa3d3c%2Fexample.png?generation=1701742942562588&amp;alt=media\" alt=\"\"></p>\n<p>And I also don't know how to generate confedence score. I took 1.0 everywhere. Can it make mistakes more drammatic in terms of scores?</p>\n<p>PS: 89% was scored on mostly unseen training data <br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F17547704%2F22f477e0482176c41c5d60c5b2efe1de%2Fexample2.png?generation=1701746152816745&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "I was analysing possible obstacles, and found out that sometimes person can be active before Wakeup is reported (applied some filters to understand the problem), in other words, the exact Onset or Wakep Events  can be not so precise... Can you explain how these events were assigned?\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F17547704%2F6a6adb6addf633688d1561a876aa3d3c%2Fexample.png?generation=1701742942562588&alt=media)\n\n\nAnd I also don't know how to generate confedence score. I took 1.0 everywhere. Can it make mistakes more drammatic in terms of scores?\n\nPS: 89% was scored on mostly unseen training data \n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F17547704%2F22f477e0482176c41c5d60c5b2efe1de%2Fexample2.png?generation=1701746152816745&alt=media)",
      "votes": null
    },
    {
      "id": "2549875",
      "postDate": "12/05/2023 15:40:57",
      "content": "<p>The metric is a bit more complicated and uses not only 30 min thresholds, but smaller ones down to 1 minute as well.</p>",
      "rawMarkdown": "The metric is a bit more complicated and uses not only 30 min thresholds, but smaller ones down to 1 minute as well.",
      "votes": null
    },
    {
      "id": "2550097",
      "postDate": "12/05/2023 19:02:16",
      "content": "<p>Thank you.</p>",
      "rawMarkdown": "Thank you.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2549875,
      "author_name": "fritzcremer",
      "author_url": "",
      "post_date": "12/05/2023 15:40:57",
      "content": "<p>The metric is a bit more complicated and uses not only 30 min thresholds, but smaller ones down to 1 minute as well.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2550097,
          "author_name": "timurgepard1988",
          "author_url": "",
          "post_date": "12/05/2023 19:02:16",
          "content": "<p>Thank you.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2549088": "I was analysing possible obstacles, and found out that sometimes person can be active before Wakeup is reported (applied some filters to understand the problem), in other words, the exact Onset or Wakep Events  can be not so precise... Can you explain how these events were assigned?\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F17547704%2F6a6adb6addf633688d1561a876aa3d3c%2Fexample.png?generation=1701742942562588&alt=media)\n\n\nAnd I also don't know how to generate confedence score. I took 1.0 everywhere. Can it make mistakes more drammatic in terms of scores?\n\nPS: 89% was scored on mostly unseen training data \n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F17547704%2F22f477e0482176c41c5d60c5b2efe1de%2Fexample2.png?generation=1701746152816745&alt=media)",
    "2549875": "The metric is a bit more complicated and uses not only 30 min thresholds, but smaller ones down to 1 minute as well.",
    "2550097": "Thank you."
  },
  "source": "meta"
}