{
  "id": 549146,
  "title": "Question about how data is being fed during inference",
  "url": "/competitions/jane-street-real-time-market-data-forecasting/discussion/549146",
  "author_name": "",
  "post_date": "2024-12-01T00:31:54.661980Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>From reading some other discussions, I'm wondering if my understanding of how the data is being fed at inference is correct:</p>\n<ol>\n<li><p>At time_id = 0 for all date_id, lags for all symbol_ids and all time_ids at date_id - 1 are provided together as one dataframe, and for time_id &gt; 0, lag is None?</p></li>\n<li><p>Data from 'test' is fed one time_id at a time sequentially from 0, so each 'test' dataframe contains data for all symbol_id for one specific date_id and time_id?</p></li>\n<li><p>The date_id in the lags corresponds to the current date_id, so 'responder_{i}_lag_1' corresponds to the target for the corresponding symbol_id and time_id from date_id - 1? Edit: does this mean that the date_id from lags (if not None) and test will always equal?</p></li>\n</ol>",
  "messages": [
    {
      "id": "3059651",
      "postDate": "12/01/2024 00:31:54",
      "content": "<p>From reading some other discussions, I'm wondering if my understanding of how the data is being fed at inference is correct:</p>\n<ol>\n<li><p>At time_id = 0 for all date_id, lags for all symbol_ids and all time_ids at date_id - 1 are provided together as one dataframe, and for time_id &gt; 0, lag is None?</p></li>\n<li><p>Data from 'test' is fed one time_id at a time sequentially from 0, so each 'test' dataframe contains data for all symbol_id for one specific date_id and time_id?</p></li>\n<li><p>The date_id in the lags corresponds to the current date_id, so 'responder_{i}_lag_1' corresponds to the target for the corresponding symbol_id and time_id from date_id - 1? Edit: does this mean that the date_id from lags (if not None) and test will always equal?</p></li>\n</ol>",
      "rawMarkdown": "From reading some other discussions, I'm wondering if my understanding of how the data is being fed at inference is correct:\n\n1. At time_id = 0 for all date_id, lags for all symbol_ids and all time_ids at date_id - 1 are provided together as one dataframe, and for time_id > 0, lag is None?\n\n2. Data from 'test' is fed one time_id at a time sequentially from 0, so each 'test' dataframe contains data for all symbol_id for one specific date_id and time_id?\n\n3. The date_id in the lags corresponds to the current date_id, so 'responder_{i}_lag_1' corresponds to the target for the corresponding symbol_id and time_id from date_id - 1? Edit: does this mean that the date_id from lags (if not None) and test will always equal?",
      "votes": null
    },
    {
      "id": "3059974",
      "postDate": "12/01/2024 09:08:48",
      "content": "<p>I would like to add one more question:</p>\n<p>4- Does the API give us complete data of a date_id for each batch? In other words, are we predicting one day at time?</p>",
      "rawMarkdown": "I would like to add one more question:\n\n4- Does the API give us complete data of a date_id for each batch? In other words, are we predicting one day at time?",
      "votes": null
    },
    {
      "id": "3060016",
      "postDate": "12/01/2024 10:25:31",
      "content": "<ol>\n<li>Yes</li>\n<li>Yes</li>\n<li>I forget, but the columns responder_{i}_lag_1 correspond to the responder values for each time_id on current_date-1</li>\n</ol>",
      "rawMarkdown": "1. Yes\n2. Yes\n3. I forget, but the columns responder_{i}_lag_1 correspond to the responder values for each time_id on current_date-1",
      "votes": null
    },
    {
      "id": "3060018",
      "postDate": "12/01/2024 10:27:14",
      "content": "<p>No, it's per time_id. That is, you respond to multiple symbols every time_id. <a href=\"https://www.kaggle.com/code/chumajin/janestreet-easy-to-understand-new-time-series-api/notebook\" target=\"_blank\">This is a great notebook on understanding the eval api.</a></p>",
      "rawMarkdown": "No, it's per time_id. That is, you respond to multiple symbols every time_id. [This is a great notebook on understanding the eval api.](https://www.kaggle.com/code/chumajin/janestreet-easy-to-understand-new-time-series-api/notebook)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3059974,
      "author_name": "serjhenrique",
      "author_url": "",
      "post_date": "12/01/2024 09:08:48",
      "content": "<p>I would like to add one more question:</p>\n<p>4- Does the API give us complete data of a date_id for each batch? In other words, are we predicting one day at time?</p>",
      "votes": null,
      "replies": [
        {
          "id": 3060018,
          "author_name": "yanisfalaki",
          "author_url": "",
          "post_date": "12/01/2024 10:27:14",
          "content": "<p>No, it's per time_id. That is, you respond to multiple symbols every time_id. <a href=\"https://www.kaggle.com/code/chumajin/janestreet-easy-to-understand-new-time-series-api/notebook\" target=\"_blank\">This is a great notebook on understanding the eval api.</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 3060016,
      "author_name": "yanisfalaki",
      "author_url": "",
      "post_date": "12/01/2024 10:25:31",
      "content": "<ol>\n<li>Yes</li>\n<li>Yes</li>\n<li>I forget, but the columns responder_{i}_lag_1 correspond to the responder values for each time_id on current_date-1</li>\n</ol>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3059651": "From reading some other discussions, I'm wondering if my understanding of how the data is being fed at inference is correct:\n\n1. At time_id = 0 for all date_id, lags for all symbol_ids and all time_ids at date_id - 1 are provided together as one dataframe, and for time_id > 0, lag is None?\n\n2. Data from 'test' is fed one time_id at a time sequentially from 0, so each 'test' dataframe contains data for all symbol_id for one specific date_id and time_id?\n\n3. The date_id in the lags corresponds to the current date_id, so 'responder_{i}_lag_1' corresponds to the target for the corresponding symbol_id and time_id from date_id - 1? Edit: does this mean that the date_id from lags (if not None) and test will always equal?",
    "3059974": "I would like to add one more question:\n\n4- Does the API give us complete data of a date_id for each batch? In other words, are we predicting one day at time?",
    "3060016": "1. Yes\n2. Yes\n3. I forget, but the columns responder_{i}_lag_1 correspond to the responder values for each time_id on current_date-1",
    "3060018": "No, it's per time_id. That is, you respond to multiple symbols every time_id. [This is a great notebook on understanding the eval api.](https://www.kaggle.com/code/chumajin/janestreet-easy-to-understand-new-time-series-api/notebook)"
  },
  "source": "meta"
}