{
  "id": 452347,
  "title": "Relevance of non-wear period prediction",
  "url": "/competitions/child-mind-institute-detect-sleep-states/discussion/452347",
  "author_name": "",
  "post_date": "2023-11-01T20:57:18.325761400Z",
  "votes": 5,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I guess many people use deep learning approaches formulating the task as a critical point detection(by fitting the Gaussian distribution around the events of interest as described in <a href=\"url\" target=\"_blank\">https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/discussion/441470</a><br>\nMy question regards mainly the inference process: the logical way to collect the potential wakeups/onsets is to choose n most probable timestamps (for both onsets and wakeups) according to the fitted model, where n - is the number of days in the series.<br>\nHowever, many series contain 'non-wear' periods, where we should output no events. Therefore, selecting n most probable events will be an overkill leading to some false positive detections. <br>\nThis way I think predicting these non-wear periods may be useful for the inference process. Has anyone already tried it?</p>",
  "messages": [
    {
      "id": "2508653",
      "postDate": "11/01/2023 20:57:18",
      "content": "<p>I guess many people use deep learning approaches formulating the task as a critical point detection(by fitting the Gaussian distribution around the events of interest as described in <a href=\"url\" target=\"_blank\">https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/discussion/441470</a><br>\nMy question regards mainly the inference process: the logical way to collect the potential wakeups/onsets is to choose n most probable timestamps (for both onsets and wakeups) according to the fitted model, where n - is the number of days in the series.<br>\nHowever, many series contain 'non-wear' periods, where we should output no events. Therefore, selecting n most probable events will be an overkill leading to some false positive detections. <br>\nThis way I think predicting these non-wear periods may be useful for the inference process. Has anyone already tried it?</p>",
      "rawMarkdown": "I guess many people use deep learning approaches formulating the task as a critical point detection(by fitting the Gaussian distribution around the events of interest as described in [https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/discussion/441470](url)\nMy question regards mainly the inference process: the logical way to collect the potential wakeups/onsets is to choose n most probable timestamps (for both onsets and wakeups) according to the fitted model, where n - is the number of days in the series.\nHowever, many series contain 'non-wear' periods, where we should output no events. Therefore, selecting n most probable events will be an overkill leading to some false positive detections. \nThis way I think predicting these non-wear periods may be useful for the inference process. Has anyone already tried it?",
      "votes": null
    },
    {
      "id": "2508684",
      "postDate": "11/01/2023 21:43:59",
      "content": "<p>Well, I didn't try yet. But if you predict the probabilities of the events, you might set the threshold for them. So in the ideal situation for the non-wear periods you should have very low probabilities of \"go to sleep\" events. </p>",
      "rawMarkdown": "Well, I didn't try yet. But if you predict the probabilities of the events, you might set the threshold for them. So in the ideal situation for the non-wear periods you should have very low probabilities of \"go to sleep\" events.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2508684,
      "author_name": "manwithaflower",
      "author_url": "",
      "post_date": "11/01/2023 21:43:59",
      "content": "<p>Well, I didn't try yet. But if you predict the probabilities of the events, you might set the threshold for them. So in the ideal situation for the non-wear periods you should have very low probabilities of \"go to sleep\" events. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2508653": "I guess many people use deep learning approaches formulating the task as a critical point detection(by fitting the Gaussian distribution around the events of interest as described in [https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states/discussion/441470](url)\nMy question regards mainly the inference process: the logical way to collect the potential wakeups/onsets is to choose n most probable timestamps (for both onsets and wakeups) according to the fitted model, where n - is the number of days in the series.\nHowever, many series contain 'non-wear' periods, where we should output no events. Therefore, selecting n most probable events will be an overkill leading to some false positive detections. \nThis way I think predicting these non-wear periods may be useful for the inference process. Has anyone already tried it?",
    "2508684": "Well, I didn't try yet. But if you predict the probabilities of the events, you might set the threshold for them. So in the ideal situation for the non-wear periods you should have very low probabilities of \"go to sleep\" events."
  },
  "source": "meta"
}