{
  "id": 553906,
  "title": "What is the forecasting horizon ?",
  "url": "/competitions/jane-street-real-time-market-data-forecasting/discussion/553906",
  "author_name": "",
  "post_date": "2024-12-29T06:13:54.088016300Z",
  "votes": -3,
  "comment_count": 3,
  "views": 0,
  "content": "<p>In the train data provided, is the responders respective to features are already forward values? Or we have to do shift(-n) &amp; what will be n  ?</p>",
  "messages": [
    {
      "id": "3083197",
      "postDate": "12/29/2024 06:13:54",
      "content": "<p>In the train data provided, is the responders respective to features are already forward values? Or we have to do shift(-n) &amp; what will be n  ?</p>",
      "rawMarkdown": "In the train data provided, is the responders respective to features are already forward values? Or we have to do shift(-n) & what will be n  ?",
      "votes": null
    },
    {
      "id": "3086929",
      "postDate": "01/02/2025 21:16:38",
      "content": "<p>The responders are likely forward values, so no need for shift(-n) unless specified. Use shift(n) to create lag features; choose n based on how far back you want to look (e.g., n=1 for the previous step). Confirm alignment in the dataset documentation.</p>",
      "rawMarkdown": "The responders are likely forward values, so no need for shift(-n) unless specified. Use shift(n) to create lag features; choose n based on how far back you want to look (e.g., n=1 for the previous step). Confirm alignment in the dataset documentation.",
      "votes": null
    },
    {
      "id": "3087150",
      "postDate": "01/03/2025 06:51:44",
      "content": "<p>Thanks for the reply, I have a doubt regarding lags :                                                                                                                                         The competition says at start time_id of current date_id we will get all values of all responders of previous date_id. Let's say the reponder_6 has an forward window of n in train data, then for this the window n &lt; difference(last time_id of prev date_id ,first time_id of curr date_id) ?  If this is True then all features are intraday data and the responders are forward data of intraday window(low window size and not extending for days) ?                                                                                                                                                                              </p>",
      "rawMarkdown": "Thanks for the reply, I have a doubt regarding lags :                                                                                                                                         The competition says at start time_id of current date_id we will get all values of all responders of previous date_id. Let's say the reponder_6 has an forward window of n in train data, then for this the window n < difference(last time_id of prev date_id ,first time_id of curr date_id) ?  If this is True then all features are intraday data and the responders are forward data of intraday window(low window size and not extending for days) ?",
      "votes": null
    },
    {
      "id": "3087304",
      "postDate": "01/03/2025 11:05:29",
      "content": "<p>What you said is correct. If the competition provides responders from the previous date_id at the start of the current date_id, then:<br>\nWindow n for responder_6 must satisfy:<br>\nn &lt; difference(last time_id of prev date_id, first time_id of curr date_id)<br>\nThis ensures responders are forward-looking but limited to intraday data.<br>\nThus, all features are intraday, and the responders are forward values within a small intraday window, not spanning multiple days.</p>",
      "rawMarkdown": "What you said is correct. If the competition provides responders from the previous date_id at the start of the current date_id, then:\nWindow n for responder_6 must satisfy:\nn < difference(last time_id of prev date_id, first time_id of curr date_id)\nThis ensures responders are forward-looking but limited to intraday data.\nThus, all features are intraday, and the responders are forward values within a small intraday window, not spanning multiple days.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3086929,
      "author_name": "rishavrajverma",
      "author_url": "",
      "post_date": "01/02/2025 21:16:38",
      "content": "<p>The responders are likely forward values, so no need for shift(-n) unless specified. Use shift(n) to create lag features; choose n based on how far back you want to look (e.g., n=1 for the previous step). Confirm alignment in the dataset documentation.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3087150,
          "author_name": "adarshmanojsingh",
          "author_url": "",
          "post_date": "01/03/2025 06:51:44",
          "content": "<p>Thanks for the reply, I have a doubt regarding lags :                                                                                                                                         The competition says at start time_id of current date_id we will get all values of all responders of previous date_id. Let's say the reponder_6 has an forward window of n in train data, then for this the window n &lt; difference(last time_id of prev date_id ,first time_id of curr date_id) ?  If this is True then all features are intraday data and the responders are forward data of intraday window(low window size and not extending for days) ?                                                                                                                                                                              </p>",
          "votes": null,
          "replies": [
            {
              "id": 3087304,
              "author_name": "rishavrajverma",
              "author_url": "",
              "post_date": "01/03/2025 11:05:29",
              "content": "<p>What you said is correct. If the competition provides responders from the previous date_id at the start of the current date_id, then:<br>\nWindow n for responder_6 must satisfy:<br>\nn &lt; difference(last time_id of prev date_id, first time_id of curr date_id)<br>\nThis ensures responders are forward-looking but limited to intraday data.<br>\nThus, all features are intraday, and the responders are forward values within a small intraday window, not spanning multiple days.</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3083197": "In the train data provided, is the responders respective to features are already forward values? Or we have to do shift(-n) & what will be n  ?",
    "3086929": "The responders are likely forward values, so no need for shift(-n) unless specified. Use shift(n) to create lag features; choose n based on how far back you want to look (e.g., n=1 for the previous step). Confirm alignment in the dataset documentation.",
    "3087150": "Thanks for the reply, I have a doubt regarding lags :                                                                                                                                         The competition says at start time_id of current date_id we will get all values of all responders of previous date_id. Let's say the reponder_6 has an forward window of n in train data, then for this the window n < difference(last time_id of prev date_id ,first time_id of curr date_id) ?  If this is True then all features are intraday data and the responders are forward data of intraday window(low window size and not extending for days) ?",
    "3087304": "What you said is correct. If the competition provides responders from the previous date_id at the start of the current date_id, then:\nWindow n for responder_6 must satisfy:\nn < difference(last time_id of prev date_id, first time_id of curr date_id)\nThis ensures responders are forward-looking but limited to intraday data.\nThus, all features are intraday, and the responders are forward values within a small intraday window, not spanning multiple days."
  },
  "source": "meta"
}