{
  "id": 396413,
  "title": "Can we use the whole sequence for each user now?",
  "url": "/competitions/predict-student-performance-from-game-play/discussion/396413",
  "author_name": "",
  "post_date": "2023-03-21T14:23:42.353098500Z",
  "votes": 6,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello.<br>\nAfter updating the API it seems that we retrieve the data for each user not level group wise but as a whole (<a href=\"https://www.kaggle.com/code/kmitsuhiro/api-update-iter-test-returns-all-test-data\" target=\"_blank\">see here</a>).<br>\nDoes that mean we can treat our problem now as the following task:<br>\nGiven a sequence for a user (x_1, … x_n) predict the corresponding labels (q_1,…,q_{18})?<br>\nFor GBT approaches in my understanding we can generate features now for the whole sequence as well as we have at inference time all the information available.<br>\nAm I right?</p>",
  "messages": [
    {
      "id": "2190838",
      "postDate": "03/21/2023 14:23:42",
      "content": "<p>Hello.<br>\nAfter updating the API it seems that we retrieve the data for each user not level group wise but as a whole (<a href=\"https://www.kaggle.com/code/kmitsuhiro/api-update-iter-test-returns-all-test-data\" target=\"_blank\">see here</a>).<br>\nDoes that mean we can treat our problem now as the following task:<br>\nGiven a sequence for a user (x_1, … x_n) predict the corresponding labels (q_1,…,q_{18})?<br>\nFor GBT approaches in my understanding we can generate features now for the whole sequence as well as we have at inference time all the information available.<br>\nAm I right?</p>",
      "rawMarkdown": "Hello.\nAfter updating the API it seems that we retrieve the data for each user not level group wise but as a whole ([see here](https://www.kaggle.com/code/kmitsuhiro/api-update-iter-test-returns-all-test-data)).\nDoes that mean we can treat our problem now as the following task:\nGiven a sequence for a user (x_1, ... x_n) predict the corresponding labels (q_1,...,q_{18})?\nFor GBT approaches in my understanding we can generate features now for the whole sequence as well as we have at inference time all the information available.\nAm I right?",
      "votes": null
    },
    {
      "id": "2191244",
      "postDate": "03/21/2023 20:32:17",
      "content": "<blockquote>\n  <p>This is an issue with the set of 3 provided public samples. The hidden test set does not have this problem</p>\n</blockquote>\n<p><a href=\"https://www.kaggle.com/competitions/predict-student-performance-from-game-play/discussion/396202#2190989\" target=\"_blank\">https://www.kaggle.com/competitions/predict-student-performance-from-game-play/discussion/396202#2190989</a></p>\n<p>The iteration in the hidden test is by group level, so you don't have all the data for a session at the same iteration</p>",
      "rawMarkdown": "> This is an issue with the set of 3 provided public samples. The hidden test set does not have this problem\n\nhttps://www.kaggle.com/competitions/predict-student-performance-from-game-play/discussion/396202#2190989\n\nThe iteration in the hidden test is by group level, so you don't have all the data for a session at the same iteration",
      "votes": null
    },
    {
      "id": "2191297",
      "postDate": "03/21/2023 21:13:16",
      "content": "<p>Tnx for the clarification!</p>",
      "rawMarkdown": "Tnx for the clarification!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2191244,
      "author_name": "steubk",
      "author_url": "",
      "post_date": "03/21/2023 20:32:17",
      "content": "<blockquote>\n  <p>This is an issue with the set of 3 provided public samples. The hidden test set does not have this problem</p>\n</blockquote>\n<p><a href=\"https://www.kaggle.com/competitions/predict-student-performance-from-game-play/discussion/396202#2190989\" target=\"_blank\">https://www.kaggle.com/competitions/predict-student-performance-from-game-play/discussion/396202#2190989</a></p>\n<p>The iteration in the hidden test is by group level, so you don't have all the data for a session at the same iteration</p>",
      "votes": null,
      "replies": [
        {
          "id": 2191297,
          "author_name": "simonveitner",
          "author_url": "",
          "post_date": "03/21/2023 21:13:16",
          "content": "<p>Tnx for the clarification!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2190838": "Hello.\nAfter updating the API it seems that we retrieve the data for each user not level group wise but as a whole ([see here](https://www.kaggle.com/code/kmitsuhiro/api-update-iter-test-returns-all-test-data)).\nDoes that mean we can treat our problem now as the following task:\nGiven a sequence for a user (x_1, ... x_n) predict the corresponding labels (q_1,...,q_{18})?\nFor GBT approaches in my understanding we can generate features now for the whole sequence as well as we have at inference time all the information available.\nAm I right?",
    "2191244": "> This is an issue with the set of 3 provided public samples. The hidden test set does not have this problem\n\nhttps://www.kaggle.com/competitions/predict-student-performance-from-game-play/discussion/396202#2190989\n\nThe iteration in the hidden test is by group level, so you don't have all the data for a session at the same iteration",
    "2191297": "Tnx for the clarification!"
  },
  "source": "meta"
}