{
  "id": 265228,
  "title": "Pseudo labeling is not effective in this competiton?",
  "url": "/competitions/seti-breakthrough-listen/discussion/265228",
  "author_name": "imori",
  "post_date": "2021-08-15T05:26:18.106000",
  "votes": 0,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hellow, everyone.</p>\n<p>I am in trouble with improving score.<br>\nI tried some models and ensemble them, but score is not improved.</p>\n<p>I tried to retrain model using test datas with pseudo labels, but result become worse….</p>\n<p>Do you use pseudo labeling in this competiton?</p>",
  "messages": [
    {
      "id": 1473292,
      "postDate": "2021-08-15T13:43:15.333Z",
      "content": "<p>I guess the big gap between train and test data is the reason why your pesudo labelling is not working, how do you fix comfirmation bias in such gap? Maybe mean teacher would help a little but I didn`t try.</p>",
      "rawMarkdown": "I guess the big gap between train and test data is the reason why your pesudo labelling is not working, how do you fix comfirmation bias in such gap? Maybe mean teacher would help a little but I didn`t try.",
      "votes": 2,
      "replies": [
        {
          "id": 1473370,
          "postDate": "2021-08-15T14:32:56.800Z",
          "content": "<p>Thank you for reply!</p>\n<blockquote>\n  <p>I guess the big gap between train and test data</p>\n</blockquote>\n<p>Yes, I think so, too.<br>\nBecause there may be gap, I thought pseudo labeling is effective approach.<br>\nHowever, when I see your opinion, I come to think pseudo label lead to worsing score.</p>\n<p>I will try other methods in last a few days, thanks!</p>",
          "rawMarkdown": "Thank you for reply!\n\n> I guess the big gap between train and test data\n\nYes, I think so, too.\nBecause there may be gap, I thought pseudo labeling is effective approach.\nHowever, when I see your opinion, I come to think pseudo label lead to worsing score.\n\nI will try other methods in last a few days, thanks!"
        }
      ]
    },
    {
      "id": 1472767,
      "postDate": "2021-08-15T05:26:18.107Z",
      "content": "<p>Hellow, everyone.</p>\n<p>I am in trouble with improving score.<br>\nI tried some models and ensemble them, but score is not improved.</p>\n<p>I tried to retrain model using test datas with pseudo labels, but result become worse….</p>\n<p>Do you use pseudo labeling in this competiton?</p>",
      "rawMarkdown": "Hellow, everyone.\n\nI am in trouble with improving score.\nI tried some models and ensemble them, but score is not improved.\n\nI tried to retrain model using test datas with pseudo labels, but result become worse....\n\nDo you use pseudo labeling in this competiton?"
    }
  ],
  "comments": [
    {
      "id": 1473292,
      "author_name": "Hao",
      "author_url": "",
      "post_date": "2021-08-15T13:43:15.333000",
      "content": "<p>I guess the big gap between train and test data is the reason why your pesudo labelling is not working, how do you fix comfirmation bias in such gap? Maybe mean teacher would help a little but I didn`t try.</p>",
      "votes": 2,
      "replies": [
        {
          "id": 1473370,
          "author_name": "imori",
          "author_url": "",
          "post_date": "2021-08-15T14:32:56.800000",
          "content": "<p>Thank you for reply!</p>\n<blockquote>\n  <p>I guess the big gap between train and test data</p>\n</blockquote>\n<p>Yes, I think so, too.<br>\nBecause there may be gap, I thought pseudo labeling is effective approach.<br>\nHowever, when I see your opinion, I come to think pseudo label lead to worsing score.</p>\n<p>I will try other methods in last a few days, thanks!</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1473292": "I guess the big gap between train and test data is the reason why your pesudo labelling is not working, how do you fix comfirmation bias in such gap? Maybe mean teacher would help a little but I didn`t try.",
    "1472767": "Hellow, everyone.\n\nI am in trouble with improving score.\nI tried some models and ensemble them, but score is not improved.\n\nI tried to retrain model using test datas with pseudo labels, but result become worse....\n\nDo you use pseudo labeling in this competiton?"
  }
}