{
  "id": 224412,
  "title": "Pseudolabelling NIH CXRs (notebook)",
  "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/224412",
  "author_name": "",
  "post_date": "2021-03-08T11:15:26.897549Z",
  "votes": 3,
  "comment_count": 2,
  "views": 0,
  "content": "<p>There's been some interesting discussion around use of the NIH data for this competition. The easiest way for a beginner like me to use the extra data is to pseudolabel it (do inference to make fake labels, so you can use the images as extra training data).</p>\n<p>I guess many of the higher teams on the leaderboard are using some variant of this approach. I thought I'd share my very rough method of doing so. Notebook is <a href=\"https://www.kaggle.com/reubenschmidt/pseudolabelling-chestx-dataset\" target=\"_blank\">here</a>. All feedback and criticisms welcome.</p>",
  "messages": [
    {
      "id": "1230691",
      "postDate": "03/08/2021 11:15:26",
      "content": "<p>There's been some interesting discussion around use of the NIH data for this competition. The easiest way for a beginner like me to use the extra data is to pseudolabel it (do inference to make fake labels, so you can use the images as extra training data).</p>\n<p>I guess many of the higher teams on the leaderboard are using some variant of this approach. I thought I'd share my very rough method of doing so. Notebook is <a href=\"https://www.kaggle.com/reubenschmidt/pseudolabelling-chestx-dataset\" target=\"_blank\">here</a>. All feedback and criticisms welcome.</p>",
      "rawMarkdown": "There's been some interesting discussion around use of the NIH data for this competition. The easiest way for a beginner like me to use the extra data is to pseudolabel it (do inference to make fake labels, so you can use the images as extra training data).\n\nI guess many of the higher teams on the leaderboard are using some variant of this approach. I thought I'd share my very rough method of doing so. Notebook is [here](https://www.kaggle.com/reubenschmidt/pseudolabelling-chestx-dataset). All feedback and criticisms welcome.",
      "votes": null
    },
    {
      "id": "1231510",
      "postDate": "03/09/2021 03:26:06",
      "content": "<p>Hi the notebook is empty. Did you commit the notebook?</p>",
      "rawMarkdown": "Hi the notebook is empty. Did you commit the notebook?",
      "votes": null
    },
    {
      "id": "1231657",
      "postDate": "03/09/2021 06:25:37",
      "content": "<p>Thanks for letting me know. Committing now</p>",
      "rawMarkdown": "Thanks for letting me know. Committing now",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1231510,
      "author_name": "reighns",
      "author_url": "",
      "post_date": "03/09/2021 03:26:06",
      "content": "<p>Hi the notebook is empty. Did you commit the notebook?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1231657,
          "author_name": "reubenschmidt",
          "author_url": "",
          "post_date": "03/09/2021 06:25:37",
          "content": "<p>Thanks for letting me know. Committing now</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1230691": "There's been some interesting discussion around use of the NIH data for this competition. The easiest way for a beginner like me to use the extra data is to pseudolabel it (do inference to make fake labels, so you can use the images as extra training data).\n\nI guess many of the higher teams on the leaderboard are using some variant of this approach. I thought I'd share my very rough method of doing so. Notebook is [here](https://www.kaggle.com/reubenschmidt/pseudolabelling-chestx-dataset). All feedback and criticisms welcome.",
    "1231510": "Hi the notebook is empty. Did you commit the notebook?",
    "1231657": "Thanks for letting me know. Committing now"
  },
  "source": "meta"
}