{
  "id": 163455,
  "title": "pseudo labeling overfit for test?",
  "url": "/competitions/global-wheat-detection/discussion/163455",
  "author_name": "",
  "post_date": "2020-07-02T05:56:45.356515400Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>We have seen many models drop like a rock during test. Do you think pseudo labeling will hold up for test?</p>\n\n<p>Thanks.</p>",
  "messages": [
    {
      "id": "911949",
      "postDate": "07/02/2020 05:56:45",
      "content": "<p>We have seen many models drop like a rock during test. Do you think pseudo labeling will hold up for test?</p>\n\n<p>Thanks.</p>",
      "rawMarkdown": "We have seen many models drop like a rock during test. Do you think pseudo labeling will hold up for test?\n\nThanks.",
      "votes": null
    },
    {
      "id": "911958",
      "postDate": "07/02/2020 06:08:16",
      "content": "<p>I think if the private test data is as easy as public data, then pseudo labeling should work. However, If the private data is more difficult, it might run into problem.</p>",
      "rawMarkdown": "I think if the private test data is as easy as public data, then pseudo labeling should work. However, If the private data is more difficult, it might run into problem.",
      "votes": null
    },
    {
      "id": "911981",
      "postDate": "07/02/2020 06:27:16",
      "content": "<p>thanks for insight </p>",
      "rawMarkdown": "thanks for insight",
      "votes": null
    },
    {
      "id": "926260",
      "postDate": "07/12/2020 15:48:19",
      "content": "<p><a href=\"/nvnnghia\">@nvnnghia</a> When we pseudo label and train it and commit.. then the weights learned from all test data??</p>",
      "rawMarkdown": "nvnnghia When we pseudo label and train it and commit.. then the weights learned from all test data??",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 911958,
      "author_name": "nvnnghia",
      "author_url": "",
      "post_date": "07/02/2020 06:08:16",
      "content": "<p>I think if the private test data is as easy as public data, then pseudo labeling should work. However, If the private data is more difficult, it might run into problem.</p>",
      "votes": null,
      "replies": [
        {
          "id": 911981,
          "author_name": "bopengiowa",
          "author_url": "",
          "post_date": "07/02/2020 06:27:16",
          "content": "<p>thanks for insight </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 926260,
          "author_name": "gyanendradas",
          "author_url": "",
          "post_date": "07/12/2020 15:48:19",
          "content": "<p><a href=\"/nvnnghia\">@nvnnghia</a> When we pseudo label and train it and commit.. then the weights learned from all test data??</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "911949": "We have seen many models drop like a rock during test. Do you think pseudo labeling will hold up for test?\n\nThanks.",
    "911958": "I think if the private test data is as easy as public data, then pseudo labeling should work. However, If the private data is more difficult, it might run into problem.",
    "911981": "thanks for insight",
    "926260": "nvnnghia When we pseudo label and train it and commit.. then the weights learned from all test data??"
  },
  "source": "meta"
}