{
  "id": 190020,
  "title": "New questions in test set",
  "url": "/competitions/riiid-test-answer-prediction/discussion/190020",
  "author_name": "redstr",
  "post_date": "2020-10-09T17:49:44.648000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>As far as I understand, there will be new, previously unseen questions in the test set. Perhaps they only appear in the private test set that will be revealed after competition is closed. Does this mean that the problem requires some kind of few-shot learning or online training in it? I checked some previous studies in knowledge tracing, and they assume a static question set, so they learn question embeddings and so on. It seems those are not directly applicable here.</p>",
  "messages": [
    {
      "id": 1044376,
      "postDate": "2020-10-09T17:49:44.650Z",
      "content": "<p>As far as I understand, there will be new, previously unseen questions in the test set. Perhaps they only appear in the private test set that will be revealed after competition is closed. Does this mean that the problem requires some kind of few-shot learning or online training in it? I checked some previous studies in knowledge tracing, and they assume a static question set, so they learn question embeddings and so on. It seems those are not directly applicable here.</p>",
      "rawMarkdown": "As far as I understand, there will be new, previously unseen questions in the test set. Perhaps they only appear in the private test set that will be revealed after competition is closed. Does this mean that the problem requires some kind of few-shot learning or online training in it? I checked some previous studies in knowledge tracing, and they assume a static question set, so they learn question embeddings and so on. It seems those are not directly applicable here.",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1044376": "As far as I understand, there will be new, previously unseen questions in the test set. Perhaps they only appear in the private test set that will be revealed after competition is closed. Does this mean that the problem requires some kind of few-shot learning or online training in it? I checked some previous studies in knowledge tracing, and they assume a static question set, so they learn question embeddings and so on. It seems those are not directly applicable here."
  }
}