{
  "id": 541912,
  "title": "The use of lags.parquet?",
  "url": "/competitions/jane-street-real-time-market-data-forecasting/discussion/541912",
  "author_name": "",
  "post_date": "2024-10-22T00:32:24.876661200Z",
  "votes": 3,
  "comment_count": 3,
  "views": 0,
  "content": "<p>lags.parquet - Values of responder_{0…8} lagged by one date_id. The evaluation API serves the entirety of the lagged responders for a date_id on that date_id's first time_id. In other words, all of the previous date's responders will be served at the first time step of the succeeding date.</p>\n<p>Anyone knows what is it for? I thought we are going to predict using the features_00 to features_78 only? Why the previous day responder?</p>\n<p>Thank you</p>",
  "messages": [
    {
      "id": "3024761",
      "postDate": "10/22/2024 00:32:24",
      "content": "<p>lags.parquet - Values of responder_{0…8} lagged by one date_id. The evaluation API serves the entirety of the lagged responders for a date_id on that date_id's first time_id. In other words, all of the previous date's responders will be served at the first time step of the succeeding date.</p>\n<p>Anyone knows what is it for? I thought we are going to predict using the features_00 to features_78 only? Why the previous day responder?</p>\n<p>Thank you</p>",
      "rawMarkdown": "lags.parquet - Values of responder_{0...8} lagged by one date_id. The evaluation API serves the entirety of the lagged responders for a date_id on that date_id's first time_id. In other words, all of the previous date's responders will be served at the first time step of the succeeding date.\n\nAnyone knows what is it for? I thought we are going to predict using the features_00 to features_78 only? Why the previous day responder?\n\nThank you",
      "votes": null
    },
    {
      "id": "3024853",
      "postDate": "10/22/2024 04:52:11",
      "content": "<p>If you are retraining your model during evaluation and other responders are part of your model inputs then this allows you to use them.</p>\n<p>Some models for example improve training on time series data that isn't strictly related. So you might find that training to predict all responders actually improves your accuracy for training on responder_6 alone.</p>",
      "rawMarkdown": "If you are retraining your model during evaluation and other responders are part of your model inputs then this allows you to use them.\n\nSome models for example improve training on time series data that isn't strictly related. So you might find that training to predict all responders actually improves your accuracy for training on responder_6 alone.",
      "votes": null
    },
    {
      "id": "3079655",
      "postDate": "12/24/2024 00:20:06",
      "content": "<p>I believe these will be essential when using online learning algorithms, since they can then adjust their weights every day to correct for mistakes they made the previous day.</p>",
      "rawMarkdown": "I believe these will be essential when using online learning algorithms, since they can then adjust their weights every day to correct for mistakes they made the previous day.",
      "votes": null
    },
    {
      "id": "3232349",
      "postDate": "06/25/2025 15:47:28",
      "content": "<p>u mean for example if i use responder_3 as ylabel to train the orignal model, it may improve the accuracy for predict responder_6?</p>",
      "rawMarkdown": "u mean for example if i use responder_3 as ylabel to train the orignal model, it may improve the accuracy for predict responder_6?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3024853,
      "author_name": "michaeltimbs",
      "author_url": "",
      "post_date": "10/22/2024 04:52:11",
      "content": "<p>If you are retraining your model during evaluation and other responders are part of your model inputs then this allows you to use them.</p>\n<p>Some models for example improve training on time series data that isn't strictly related. So you might find that training to predict all responders actually improves your accuracy for training on responder_6 alone.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3232349,
          "author_name": "louyukun",
          "author_url": "",
          "post_date": "06/25/2025 15:47:28",
          "content": "<p>u mean for example if i use responder_3 as ylabel to train the orignal model, it may improve the accuracy for predict responder_6?</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 3079655,
      "author_name": "darkonyx",
      "author_url": "",
      "post_date": "12/24/2024 00:20:06",
      "content": "<p>I believe these will be essential when using online learning algorithms, since they can then adjust their weights every day to correct for mistakes they made the previous day.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3024761": "lags.parquet - Values of responder_{0...8} lagged by one date_id. The evaluation API serves the entirety of the lagged responders for a date_id on that date_id's first time_id. In other words, all of the previous date's responders will be served at the first time step of the succeeding date.\n\nAnyone knows what is it for? I thought we are going to predict using the features_00 to features_78 only? Why the previous day responder?\n\nThank you",
    "3024853": "If you are retraining your model during evaluation and other responders are part of your model inputs then this allows you to use them.\n\nSome models for example improve training on time series data that isn't strictly related. So you might find that training to predict all responders actually improves your accuracy for training on responder_6 alone.",
    "3079655": "I believe these will be essential when using online learning algorithms, since they can then adjust their weights every day to correct for mistakes they made the previous day.",
    "3232349": "u mean for example if i use responder_3 as ylabel to train the orignal model, it may improve the accuracy for predict responder_6?"
  },
  "source": "meta"
}