{
  "id": 459107,
  "title": "High level RNN approach ",
  "url": "/competitions/child-mind-institute-detect-sleep-states/discussion/459107",
  "author_name": "Max Rivera",
  "post_date": "2023-12-03T15:45:13.943000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi guys, trying to understand how a high level RNN approach would work for this kind of problem. One way could be to divide the time series up into windows and do a many-to-one prediction for each window, so every time step in that window gets the same label (less precise). Or do a many-to-many prediction on each window (more computationally expensive?). And if you downsample you would have to assign the same label for everything in that sample. Any thoughts? </p>",
  "messages": [
    {
      "id": 2547576,
      "postDate": "2023-12-03T15:45:13.943Z",
      "content": "<p>Hi guys, trying to understand how a high level RNN approach would work for this kind of problem. One way could be to divide the time series up into windows and do a many-to-one prediction for each window, so every time step in that window gets the same label (less precise). Or do a many-to-many prediction on each window (more computationally expensive?). And if you downsample you would have to assign the same label for everything in that sample. Any thoughts? </p>",
      "rawMarkdown": "Hi guys, trying to understand how a high level RNN approach would work for this kind of problem. One way could be to divide the time series up into windows and do a many-to-one prediction for each window, so every time step in that window gets the same label (less precise). Or do a many-to-many prediction on each window (more computationally expensive?). And if you downsample you would have to assign the same label for everything in that sample. Any thoughts? "
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2547576": "Hi guys, trying to understand how a high level RNN approach would work for this kind of problem. One way could be to divide the time series up into windows and do a many-to-one prediction for each window, so every time step in that window gets the same label (less precise). Or do a many-to-many prediction on each window (more computationally expensive?). And if you downsample you would have to assign the same label for everything in that sample. Any thoughts? "
  }
}