{
  "id": 398984,
  "title": "Using features about past actions.",
  "url": "/competitions/predict-student-performance-from-game-play/discussion/398984",
  "author_name": "",
  "post_date": "2023-04-01T20:16:24.394484600Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello everyone. I have a question regarding features. I decided to try using information about past actions, such as how much time people spent answering questions after a group of levels. After generating these feature data, my CV got a decent boost, but the score on the hidden test significantly decreased. As a beginner, I would like to ask more experienced individuals what could be the cause of this. Thank you in advance.</p>",
  "messages": [
    {
      "id": "2205714",
      "postDate": "04/01/2023 20:16:24",
      "content": "<p>Hello everyone. I have a question regarding features. I decided to try using information about past actions, such as how much time people spent answering questions after a group of levels. After generating these feature data, my CV got a decent boost, but the score on the hidden test significantly decreased. As a beginner, I would like to ask more experienced individuals what could be the cause of this. Thank you in advance.</p>",
      "rawMarkdown": "Hello everyone. I have a question regarding features. I decided to try using information about past actions, such as how much time people spent answering questions after a group of levels. After generating these feature data, my CV got a decent boost, but the score on the hidden test significantly decreased. As a beginner, I would like to ask more experienced individuals what could be the cause of this. Thank you in advance.",
      "votes": null
    },
    {
      "id": "2208385",
      "postDate": "04/04/2023 02:30:03",
      "content": "<p>Using the time answering questions can boost CV by <code>0.010</code> but during inference we <strong>cannot</strong> get this time. Because the Kaggle API gives us levels 0-4 to predict questions 1-3. Then the API gives us levels 5-12 to predict questions 4-13. Then the API gives us levels 13-22 to predict questions 14-18. So when we predict questions 1-3 we <strong>do not</strong> have the time between level 3 and level 4 which is the time taken to answer questions. And when predicting questions 4-13, we <strong>do not</strong> have the time between level 13 and 14 which is the time take to answer questions etc etc</p>",
      "rawMarkdown": "Using the time answering questions can boost CV by `0.010` but during inference we **cannot** get this time. Because the Kaggle API gives us levels 0-4 to predict questions 1-3. Then the API gives us levels 5-12 to predict questions 4-13. Then the API gives us levels 13-22 to predict questions 14-18. So when we predict questions 1-3 we **do not** have the time between level 3 and level 4 which is the time taken to answer questions. And when predicting questions 4-13, we **do not** have the time between level 13 and 14 which is the time take to answer questions etc etc",
      "votes": null
    },
    {
      "id": "2208398",
      "postDate": "04/04/2023 03:02:10",
      "content": "<p>Thank you for your response. </p>",
      "rawMarkdown": "Thank you for your response.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2208385,
      "author_name": "cdeotte",
      "author_url": "",
      "post_date": "04/04/2023 02:30:03",
      "content": "<p>Using the time answering questions can boost CV by <code>0.010</code> but during inference we <strong>cannot</strong> get this time. Because the Kaggle API gives us levels 0-4 to predict questions 1-3. Then the API gives us levels 5-12 to predict questions 4-13. Then the API gives us levels 13-22 to predict questions 14-18. So when we predict questions 1-3 we <strong>do not</strong> have the time between level 3 and level 4 which is the time taken to answer questions. And when predicting questions 4-13, we <strong>do not</strong> have the time between level 13 and 14 which is the time take to answer questions etc etc</p>",
      "votes": null,
      "replies": [
        {
          "id": 2208398,
          "author_name": "thedark735",
          "author_url": "",
          "post_date": "04/04/2023 03:02:10",
          "content": "<p>Thank you for your response. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2205714": "Hello everyone. I have a question regarding features. I decided to try using information about past actions, such as how much time people spent answering questions after a group of levels. After generating these feature data, my CV got a decent boost, but the score on the hidden test significantly decreased. As a beginner, I would like to ask more experienced individuals what could be the cause of this. Thank you in advance.",
    "2208385": "Using the time answering questions can boost CV by `0.010` but during inference we **cannot** get this time. Because the Kaggle API gives us levels 0-4 to predict questions 1-3. Then the API gives us levels 5-12 to predict questions 4-13. Then the API gives us levels 13-22 to predict questions 14-18. So when we predict questions 1-3 we **do not** have the time between level 3 and level 4 which is the time taken to answer questions. And when predicting questions 4-13, we **do not** have the time between level 13 and 14 which is the time take to answer questions etc etc",
    "2208398": "Thank you for your response."
  },
  "source": "meta"
}