{
  "id": 209873,
  "title": "Did anyone manage to build a solution involving re-training of a model during inference?",
  "url": "/competitions/riiid-test-answer-prediction/discussion/209873",
  "author_name": "",
  "post_date": "2021-01-08T22:09:38.293189Z",
  "votes": 4,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Or was that not feasible at all for this type of competition?</p>",
  "messages": [
    {
      "id": "1145098",
      "postDate": "01/08/2021 22:09:38",
      "content": "<p>Or was that not feasible at all for this type of competition?</p>",
      "rawMarkdown": "Or was that not feasible at all for this type of competition?",
      "votes": null
    },
    {
      "id": "1148228",
      "postDate": "01/11/2021 02:45:12",
      "content": "<p>Hi Nya: retraining during inference is routine in actual teach-and-test situations. However, the models are much simpler since they must operate in real time, separately for each student, and also select the next question/feedback/lecture to administer. I used such a model and you can see where it got me on the leaderboard (2100) ! I was around 70% accurate. The winner was 75% accurate. Is this difference important in practice? I don't know. Perhaps Riiid! can tell us.</p>",
      "rawMarkdown": "Hi Nya: retraining during inference is routine in actual teach-and-test situations. However, the models are much simpler since they must operate in real time, separately for each student, and also select the next question/feedback/lecture to administer. I used such a model and you can see where it got me on the leaderboard (2100) ! I was around 70% accurate. The winner was 75% accurate. Is this difference important in practice? I don't know. Perhaps Riiid! can tell us.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1148228,
      "author_name": "mikel1",
      "author_url": "",
      "post_date": "01/11/2021 02:45:12",
      "content": "<p>Hi Nya: retraining during inference is routine in actual teach-and-test situations. However, the models are much simpler since they must operate in real time, separately for each student, and also select the next question/feedback/lecture to administer. I used such a model and you can see where it got me on the leaderboard (2100) ! I was around 70% accurate. The winner was 75% accurate. Is this difference important in practice? I don't know. Perhaps Riiid! can tell us.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1145098": "Or was that not feasible at all for this type of competition?",
    "1148228": "Hi Nya: retraining during inference is routine in actual teach-and-test situations. However, the models are much simpler since they must operate in real time, separately for each student, and also select the next question/feedback/lecture to administer. I used such a model and you can see where it got me on the leaderboard (2100) ! I was around 70% accurate. The winner was 75% accurate. Is this difference important in practice? I don't know. Perhaps Riiid! can tell us."
  },
  "source": "meta"
}