{
  "id": 542897,
  "title": "Availability of features during inference ",
  "url": "/competitions/jane-street-real-time-market-data-forecasting/discussion/542897",
  "author_name": "",
  "post_date": "2024-10-27T13:31:18.290875800Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>During the evaluation, are the features (not responders) at every time point available for online learning? How can we access them in our code?</p>",
  "messages": [
    {
      "id": "3029585",
      "postDate": "10/27/2024 13:31:18",
      "content": "<p>During the evaluation, are the features (not responders) at every time point available for online learning? How can we access them in our code?</p>",
      "rawMarkdown": "During the evaluation, are the features (not responders) at every time point available for online learning? How can we access them in our code?",
      "votes": null
    },
    {
      "id": "3029598",
      "postDate": "10/27/2024 13:45:33",
      "content": "<p><a href=\"https://www.kaggle.com/competitions/jane-street-real-time-market-data-forecasting/data\" target=\"_blank\">https://www.kaggle.com/competitions/jane-street-real-time-market-data-forecasting/data</a> <a href=\"https://www.kaggle.com/chaotic\" target=\"_blank\">@chaotic</a> </p>\n<p>Kindly read this page for more details<br>\nYou will receive all features by <strong>date_id</strong> and <strong>time_id</strong> in batches. So at one time, you will get 35-40 rows of data/ When time_id == 0, you will het the lagged values of all previous day's targets as well. </p>",
      "rawMarkdown": "https://www.kaggle.com/competitions/jane-street-real-time-market-data-forecasting/data @chaotic \n\nKindly read this page for more details\nYou will receive all features by **date_id** and **time_id** in batches. So at one time, you will get 35-40 rows of data/ When time_id == 0, you will het the lagged values of all previous day's targets as well.",
      "votes": null
    },
    {
      "id": "3030368",
      "postDate": "10/28/2024 13:10:02",
      "content": "<p>So, we can create a global variable ourselves to store the previous day's test data and update it based on time_id=0?</p>",
      "rawMarkdown": "So, we can create a global variable ourselves to store the previous day's test data and update it based on time_id=0?",
      "votes": null
    },
    {
      "id": "3030389",
      "postDate": "10/28/2024 13:29:19",
      "content": "<p>yes. You can check my kernel <a href=\"https://www.kaggle.com/code/shiyili/js2024-rmf-gru-inference\" target=\"_blank\"><strong>here</strong></a> as an example. </p>\n<p>Note that it is not trivial to manage the historical cahce with a good iteration speed. How to do it properly is heavily dependent on your input design.</p>",
      "rawMarkdown": "yes. You can check my kernel [**here**](https://www.kaggle.com/code/shiyili/js2024-rmf-gru-inference) as an example. \n\nNote that it is not trivial to manage the historical cahce with a good iteration speed. How to do it properly is heavily dependent on your input design.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3029598,
      "author_name": "ravi20076",
      "author_url": "",
      "post_date": "10/27/2024 13:45:33",
      "content": "<p><a href=\"https://www.kaggle.com/competitions/jane-street-real-time-market-data-forecasting/data\" target=\"_blank\">https://www.kaggle.com/competitions/jane-street-real-time-market-data-forecasting/data</a> <a href=\"https://www.kaggle.com/chaotic\" target=\"_blank\">@chaotic</a> </p>\n<p>Kindly read this page for more details<br>\nYou will receive all features by <strong>date_id</strong> and <strong>time_id</strong> in batches. So at one time, you will get 35-40 rows of data/ When time_id == 0, you will het the lagged values of all previous day's targets as well. </p>",
      "votes": null,
      "replies": [
        {
          "id": 3030368,
          "author_name": "yunsuxiaozi",
          "author_url": "",
          "post_date": "10/28/2024 13:10:02",
          "content": "<p>So, we can create a global variable ourselves to store the previous day's test data and update it based on time_id=0?</p>",
          "votes": null,
          "replies": [
            {
              "id": 3030389,
              "author_name": "shiyili",
              "author_url": "",
              "post_date": "10/28/2024 13:29:19",
              "content": "<p>yes. You can check my kernel <a href=\"https://www.kaggle.com/code/shiyili/js2024-rmf-gru-inference\" target=\"_blank\"><strong>here</strong></a> as an example. </p>\n<p>Note that it is not trivial to manage the historical cahce with a good iteration speed. How to do it properly is heavily dependent on your input design.</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3029585": "During the evaluation, are the features (not responders) at every time point available for online learning? How can we access them in our code?",
    "3029598": "https://www.kaggle.com/competitions/jane-street-real-time-market-data-forecasting/data @chaotic \n\nKindly read this page for more details\nYou will receive all features by **date_id** and **time_id** in batches. So at one time, you will get 35-40 rows of data/ When time_id == 0, you will het the lagged values of all previous day's targets as well.",
    "3030368": "So, we can create a global variable ourselves to store the previous day's test data and update it based on time_id=0?",
    "3030389": "yes. You can check my kernel [**here**](https://www.kaggle.com/code/shiyili/js2024-rmf-gru-inference) as an example. \n\nNote that it is not trivial to manage the historical cahce with a good iteration speed. How to do it properly is heavily dependent on your input design."
  },
  "source": "meta"
}