{
  "id": 443382,
  "title": "Please give me your advice about predicting the occurrence of events using time-series data!",
  "url": "/competitions/child-mind-institute-detect-sleep-states/discussion/443382",
  "author_name": "",
  "post_date": "2023-09-27T01:41:28.786542800Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>It has been a week since I participated in this competition.<br>\nLooking at your code, it seems to me that you are predicting the occurrence of events without regard to the timeline…<br>\nI think you have a point, but I am wondering if it is possible to predict the occurrence of events using time-series data!<br>\nI am not familiar with time series data, so I would like to borrow your wisdom! <br>\nPlease give me your advice!</p>\n<hr>\n<p>My idea is to break the current data into several pieces of data and flag the occurrence of events to accompany them.<br>\nFor example, if the original data is divided into 1000 data and 500 of them are labeled wakeup(flag:0) and the rest are labeled onset(flag:1), I would like to extract 1000 data and flag the flag as 0.5 event occurrence flag by taking the average. I wonder if it would be a good idea….</p>",
  "messages": [
    {
      "id": "2457500",
      "postDate": "09/27/2023 01:41:28",
      "content": "<p>It has been a week since I participated in this competition.<br>\nLooking at your code, it seems to me that you are predicting the occurrence of events without regard to the timeline…<br>\nI think you have a point, but I am wondering if it is possible to predict the occurrence of events using time-series data!<br>\nI am not familiar with time series data, so I would like to borrow your wisdom! <br>\nPlease give me your advice!</p>\n<hr>\n<p>My idea is to break the current data into several pieces of data and flag the occurrence of events to accompany them.<br>\nFor example, if the original data is divided into 1000 data and 500 of them are labeled wakeup(flag:0) and the rest are labeled onset(flag:1), I would like to extract 1000 data and flag the flag as 0.5 event occurrence flag by taking the average. I wonder if it would be a good idea….</p>",
      "rawMarkdown": "It has been a week since I participated in this competition.\nLooking at your code, it seems to me that you are predicting the occurrence of events without regard to the timeline...\nI think you have a point, but I am wondering if it is possible to predict the occurrence of events using time-series data!\nI am not familiar with time series data, so I would like to borrow your wisdom! \nPlease give me your advice!\n\n-----------------------------------------\nMy idea is to break the current data into several pieces of data and flag the occurrence of events to accompany them.\nFor example, if the original data is divided into 1000 data and 500 of them are labeled wakeup(flag:0) and the rest are labeled onset(flag:1), I would like to extract 1000 data and flag the flag as 0.5 event occurrence flag by taking the average. I wonder if it would be a good idea....",
      "votes": null
    },
    {
      "id": "2457518",
      "postDate": "09/27/2023 02:28:26",
      "content": "<p>Summary from public notebooks on how to infer events based on prediction score:</p>\n<ol>\n<li>Smooth the prediction score from few previous time steps</li>\n<li>Set score on each step to 1 if it is higher than or equal to a threshold, otherwise set to 0 (mostly one will set threshold to 0.5, which is basically this step is rounding to nearest integer)</li>\n<li>Based on the result in step 2, an event is timesteps which has consecutive 1's</li>\n</ol>",
      "rawMarkdown": "Summary from public notebooks on how to infer events based on prediction score:\n\n1. Smooth the prediction score from few previous time steps\n2. Set score on each step to 1 if it is higher than or equal to a threshold, otherwise set to 0 (mostly one will set threshold to 0.5, which is basically this step is rounding to nearest integer)\n3. Based on the result in step 2, an event is timesteps which has consecutive 1's",
      "votes": null
    },
    {
      "id": "2457542",
      "postDate": "09/27/2023 02:59:33",
      "content": "<p>Thanks for your comment!</p>\n<p>If you don't mind, could you please share the URL of that public notebook?</p>",
      "rawMarkdown": "Thanks for your comment!\n\nIf you don't mind, could you please share the URL of that public notebook?",
      "votes": null
    },
    {
      "id": "2457567",
      "postDate": "09/27/2023 03:54:07",
      "content": "<p>Basically most of the public notebooks use this scheme, but this is one with the most votes: <a href=\"https://www.kaggle.com/code/carlmcbrideellis/zzzs-random-forest-model-starter?scriptVersionId=142971069&amp;cellId=12\" target=\"_blank\">LINK</a></p>",
      "rawMarkdown": "Basically most of the public notebooks use this scheme, but this is one with the most votes: [LINK](https://www.kaggle.com/code/carlmcbrideellis/zzzs-random-forest-model-starter?scriptVersionId=142971069&cellId=12)",
      "votes": null
    },
    {
      "id": "2457623",
      "postDate": "09/27/2023 04:56:13",
      "content": "<p>Thanks for letting me know!☺️<br>\nI'll give it a try!</p>\n<p>Good luck to both of us!</p>",
      "rawMarkdown": "Thanks for letting me know!☺️\nI'll give it a try!\n\nGood luck to both of us!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2457518,
      "author_name": "rimbax",
      "author_url": "",
      "post_date": "09/27/2023 02:28:26",
      "content": "<p>Summary from public notebooks on how to infer events based on prediction score:</p>\n<ol>\n<li>Smooth the prediction score from few previous time steps</li>\n<li>Set score on each step to 1 if it is higher than or equal to a threshold, otherwise set to 0 (mostly one will set threshold to 0.5, which is basically this step is rounding to nearest integer)</li>\n<li>Based on the result in step 2, an event is timesteps which has consecutive 1's</li>\n</ol>",
      "votes": null,
      "replies": [
        {
          "id": 2457542,
          "author_name": "tramikan",
          "author_url": "",
          "post_date": "09/27/2023 02:59:33",
          "content": "<p>Thanks for your comment!</p>\n<p>If you don't mind, could you please share the URL of that public notebook?</p>",
          "votes": null,
          "replies": [
            {
              "id": 2457567,
              "author_name": "rimbax",
              "author_url": "",
              "post_date": "09/27/2023 03:54:07",
              "content": "<p>Basically most of the public notebooks use this scheme, but this is one with the most votes: <a href=\"https://www.kaggle.com/code/carlmcbrideellis/zzzs-random-forest-model-starter?scriptVersionId=142971069&amp;cellId=12\" target=\"_blank\">LINK</a></p>",
              "votes": null,
              "replies": [
                {
                  "id": 2457623,
                  "author_name": "tramikan",
                  "author_url": "",
                  "post_date": "09/27/2023 04:56:13",
                  "content": "<p>Thanks for letting me know!☺️<br>\nI'll give it a try!</p>\n<p>Good luck to both of us!</p>",
                  "votes": null,
                  "replies": []
                }
              ]
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2457500": "It has been a week since I participated in this competition.\nLooking at your code, it seems to me that you are predicting the occurrence of events without regard to the timeline...\nI think you have a point, but I am wondering if it is possible to predict the occurrence of events using time-series data!\nI am not familiar with time series data, so I would like to borrow your wisdom! \nPlease give me your advice!\n\n-----------------------------------------\nMy idea is to break the current data into several pieces of data and flag the occurrence of events to accompany them.\nFor example, if the original data is divided into 1000 data and 500 of them are labeled wakeup(flag:0) and the rest are labeled onset(flag:1), I would like to extract 1000 data and flag the flag as 0.5 event occurrence flag by taking the average. I wonder if it would be a good idea....",
    "2457518": "Summary from public notebooks on how to infer events based on prediction score:\n\n1. Smooth the prediction score from few previous time steps\n2. Set score on each step to 1 if it is higher than or equal to a threshold, otherwise set to 0 (mostly one will set threshold to 0.5, which is basically this step is rounding to nearest integer)\n3. Based on the result in step 2, an event is timesteps which has consecutive 1's",
    "2457542": "Thanks for your comment!\n\nIf you don't mind, could you please share the URL of that public notebook?",
    "2457567": "Basically most of the public notebooks use this scheme, but this is one with the most votes: [LINK](https://www.kaggle.com/code/carlmcbrideellis/zzzs-random-forest-model-starter?scriptVersionId=142971069&cellId=12)",
    "2457623": "Thanks for letting me know!☺️\nI'll give it a try!\n\nGood luck to both of us!"
  },
  "source": "meta"
}