{
  "id": 202513,
  "title": "Would Pseudo-Labelling help in this competition?",
  "url": "/competitions/riiid-test-answer-prediction/discussion/202513",
  "author_name": "",
  "post_date": "2020-12-10T13:47:54.292712800Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I recently learnt about the concept of pseudo-labelling from this amazing <a href=\"https://www.youtube.com/watch?v=SsnWM1xWDu4\" target=\"_blank\">video</a>. My question is would it help in this competition because the public test set is pretty small.<br>\nThanks :)</p>",
  "messages": [
    {
      "id": "1108276",
      "postDate": "12/10/2020 13:47:54",
      "content": "<p>I recently learnt about the concept of pseudo-labelling from this amazing <a href=\"https://www.youtube.com/watch?v=SsnWM1xWDu4\" target=\"_blank\">video</a>. My question is would it help in this competition because the public test set is pretty small.<br>\nThanks :)</p>",
      "rawMarkdown": "I recently learnt about the concept of pseudo-labelling from this amazing [video](https://www.youtube.com/watch?v=SsnWM1xWDu4). My question is would it help in this competition because the public test set is pretty small.\nThanks :)",
      "votes": null
    },
    {
      "id": "1108366",
      "postDate": "12/10/2020 15:21:38",
      "content": "<p>Pseudo labelling is very useful when we have comparable amount of unlabelled data to labelled data. This competition is very bad for using Pseudo labelling. The amount of test data available is very less ~2.5M compared to 100M train data. You can still try to do pseudo labelling but the results might not be promising (It's just my insight from my previous experiences). </p>",
      "rawMarkdown": "Pseudo labelling is very useful when we have comparable amount of unlabelled data to labelled data. This competition is very bad for using Pseudo labelling. The amount of test data available is very less ~2.5M compared to 100M train data. You can still try to do pseudo labelling but the results might not be promising (It's just my insight from my previous experiences).",
      "votes": null
    },
    {
      "id": "1108383",
      "postDate": "12/10/2020 15:35:58",
      "content": "<p>Thanks a lot!</p>",
      "rawMarkdown": "Thanks a lot!",
      "votes": null
    },
    {
      "id": "1108902",
      "postDate": "12/11/2020 06:04:07",
      "content": "<p>I really doubt it</p>",
      "rawMarkdown": "I really doubt it",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1108366,
      "author_name": "manikanthr5",
      "author_url": "",
      "post_date": "12/10/2020 15:21:38",
      "content": "<p>Pseudo labelling is very useful when we have comparable amount of unlabelled data to labelled data. This competition is very bad for using Pseudo labelling. The amount of test data available is very less ~2.5M compared to 100M train data. You can still try to do pseudo labelling but the results might not be promising (It's just my insight from my previous experiences). </p>",
      "votes": null,
      "replies": [
        {
          "id": 1108383,
          "author_name": "aryankhatana",
          "author_url": "",
          "post_date": "12/10/2020 15:35:58",
          "content": "<p>Thanks a lot!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1108902,
      "author_name": "stecasasso",
      "author_url": "",
      "post_date": "12/11/2020 06:04:07",
      "content": "<p>I really doubt it</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1108276": "I recently learnt about the concept of pseudo-labelling from this amazing [video](https://www.youtube.com/watch?v=SsnWM1xWDu4). My question is would it help in this competition because the public test set is pretty small.\nThanks :)",
    "1108366": "Pseudo labelling is very useful when we have comparable amount of unlabelled data to labelled data. This competition is very bad for using Pseudo labelling. The amount of test data available is very less ~2.5M compared to 100M train data. You can still try to do pseudo labelling but the results might not be promising (It's just my insight from my previous experiences).",
    "1108383": "Thanks a lot!",
    "1108902": "I really doubt it"
  },
  "source": "meta"
}