{
  "id": 438811,
  "title": "Predicting state vs event",
  "url": "/competitions/child-mind-institute-detect-sleep-states/discussion/438811",
  "author_name": "",
  "post_date": "2023-09-12T16:36:44.756964500Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I haven't really worked with time series data in the past. People seem to be predicting state (awake vs asleep), then converting that into events (wakeup, onset).</p>\n<p>Is this standard practice over predicting events directly?</p>\n<p>I guess predicting events would be incredibly unbalanced, but hoping someone with more experience could chime in!</p>",
  "messages": [
    {
      "id": "2435007",
      "postDate": "09/12/2023 16:36:44",
      "content": "<p>I haven't really worked with time series data in the past. People seem to be predicting state (awake vs asleep), then converting that into events (wakeup, onset).</p>\n<p>Is this standard practice over predicting events directly?</p>\n<p>I guess predicting events would be incredibly unbalanced, but hoping someone with more experience could chime in!</p>",
      "rawMarkdown": "I haven't really worked with time series data in the past. People seem to be predicting state (awake vs asleep), then converting that into events (wakeup, onset).\n\nIs this standard practice over predicting events directly?\n\nI guess predicting events would be incredibly unbalanced, but hoping someone with more experience could chime in!",
      "votes": null
    },
    {
      "id": "2435275",
      "postDate": "09/12/2023 20:41:09",
      "content": "<p>I am having this some question myself before I continue with my attempt at this competition.</p>\n<p>I presume you get a much more balanced dataset if you use 'asleep' as the target than if you use the events ('onset', 'wakeup') and that is good. </p>\n<p>On the down side, you have many more samples to use to train your model than if you just want it to learn to predict the transitions (the events). And that is bad.</p>",
      "rawMarkdown": "I am having this some question myself before I continue with my attempt at this competition.\n\nI presume you get a much more balanced dataset if you use 'asleep' as the target than if you use the events ('onset', 'wakeup') and that is good. \n\nOn the down side, you have many more samples to use to train your model than if you just want it to learn to predict the transitions (the events). And that is bad.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2435275,
      "author_name": "aiaiaidavid",
      "author_url": "",
      "post_date": "09/12/2023 20:41:09",
      "content": "<p>I am having this some question myself before I continue with my attempt at this competition.</p>\n<p>I presume you get a much more balanced dataset if you use 'asleep' as the target than if you use the events ('onset', 'wakeup') and that is good. </p>\n<p>On the down side, you have many more samples to use to train your model than if you just want it to learn to predict the transitions (the events). And that is bad.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2435007": "I haven't really worked with time series data in the past. People seem to be predicting state (awake vs asleep), then converting that into events (wakeup, onset).\n\nIs this standard practice over predicting events directly?\n\nI guess predicting events would be incredibly unbalanced, but hoping someone with more experience could chime in!",
    "2435275": "I am having this some question myself before I continue with my attempt at this competition.\n\nI presume you get a much more balanced dataset if you use 'asleep' as the target than if you use the events ('onset', 'wakeup') and that is good. \n\nOn the down side, you have many more samples to use to train your model than if you just want it to learn to predict the transitions (the events). And that is bad."
  },
  "source": "meta"
}