{
  "id": 544951,
  "title": "Model the problem as sequence to value",
  "url": "/competitions/jane-street-real-time-market-data-forecasting/discussion/544951",
  "author_name": "",
  "post_date": "2024-11-07T17:32:13.120390900Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi guys, i am a bit confused, the data are time-series but everybody use tree-based models. Is this because with the api in inference we cannot utilize past rows as input to predict the target? Is the intention of the host to only use the row features to predict the target? but then i saw some notebooks that use lags which is data from previous rows.</p>",
  "messages": [
    {
      "id": "3039134",
      "postDate": "11/07/2024 17:32:13",
      "content": "<p>Hi guys, i am a bit confused, the data are time-series but everybody use tree-based models. Is this because with the api in inference we cannot utilize past rows as input to predict the target? Is the intention of the host to only use the row features to predict the target? but then i saw some notebooks that use lags which is data from previous rows.</p>",
      "rawMarkdown": "Hi guys, i am a bit confused, the data are time-series but everybody use tree-based models. Is this because with the api in inference we cannot utilize past rows as input to predict the target? Is the intention of the host to only use the row features to predict the target? but then i saw some notebooks that use lags which is data from previous rows.",
      "votes": null
    },
    {
      "id": "3039341",
      "postDate": "11/07/2024 23:55:31",
      "content": "<p>Yes you can use sequence models by manually storing past test dataframes provided by the api</p>",
      "rawMarkdown": "Yes you can use sequence models by manually storing past test dataframes provided by the api",
      "votes": null
    },
    {
      "id": "3039403",
      "postDate": "11/08/2024 02:48:59",
      "content": "<p>I personally think that establishing an appropriate time step is too difficult because the date_id, time_id, and symbol_id are not consistently continuous.</p>",
      "rawMarkdown": "I personally think that establishing an appropriate time step is too difficult because the date_id, time_id, and symbol_id are not consistently continuous.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3039341,
      "author_name": "snehalverma10",
      "author_url": "",
      "post_date": "11/07/2024 23:55:31",
      "content": "<p>Yes you can use sequence models by manually storing past test dataframes provided by the api</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3039403,
      "author_name": "swufeleo",
      "author_url": "",
      "post_date": "11/08/2024 02:48:59",
      "content": "<p>I personally think that establishing an appropriate time step is too difficult because the date_id, time_id, and symbol_id are not consistently continuous.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3039134": "Hi guys, i am a bit confused, the data are time-series but everybody use tree-based models. Is this because with the api in inference we cannot utilize past rows as input to predict the target? Is the intention of the host to only use the row features to predict the target? but then i saw some notebooks that use lags which is data from previous rows.",
    "3039341": "Yes you can use sequence models by manually storing past test dataframes provided by the api",
    "3039403": "I personally think that establishing an appropriate time step is too difficult because the date_id, time_id, and symbol_id are not consistently continuous."
  },
  "source": "meta"
}