{
  "id": 76280,
  "title": "is hpa external data \"really clean\" ?",
  "url": "/competitions/human-protein-atlas-image-classification/discussion/76280",
  "author_name": "hengck23",
  "post_date": "2018-12-31T14:22:29.560000",
  "votes": 8,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Is there a difference between the kaggle and the hpa external data?</p>\n\n<ol>\n<li><p>there seem to be a difference in the intensity,  especially the green channels.</p></li>\n<li><p>hpa label can be classified as Enhanced, Supported, Approved, Uncertain? It is unclear that how kaggle data is labelled?</p></li>\n</ol>\n\n<p>did anyone make some comparison of the followings:</p>\n\n<ol>\n<li>duplicates are affecting public LB results (change of distribution) or not</li>\n<li>effects of using kaggle only,  kaggle+external_hpa, external_hpa only,  kaggle+external_hpa, but finetune on hpa</li>\n<li>if addition of external_hpa makes training/validation unstable</li>\n</ol>\n\n<p>Thanks.</p>\n\n<p><a href=\"https://www.proteinatlas.org/about/assays+annotation\">https://www.proteinatlas.org/about/assays+annotation</a></p>",
  "messages": [
    {
      "id": 448225,
      "postDate": "2018-12-31T14:22:29.560Z",
      "content": "<p>Is there a difference between the kaggle and the hpa external data?</p>\n\n<ol>\n<li><p>there seem to be a difference in the intensity,  especially the green channels.</p></li>\n<li><p>hpa label can be classified as Enhanced, Supported, Approved, Uncertain? It is unclear that how kaggle data is labelled?</p></li>\n</ol>\n\n<p>did anyone make some comparison of the followings:</p>\n\n<ol>\n<li>duplicates are affecting public LB results (change of distribution) or not</li>\n<li>effects of using kaggle only,  kaggle+external_hpa, external_hpa only,  kaggle+external_hpa, but finetune on hpa</li>\n<li>if addition of external_hpa makes training/validation unstable</li>\n</ol>\n\n<p>Thanks.</p>\n\n<p><a href=\"https://www.proteinatlas.org/about/assays+annotation\">https://www.proteinatlas.org/about/assays+annotation</a></p>",
      "rawMarkdown": "Is there a difference between the kaggle and the hpa external data?\n\n 1.  there seem to be a difference in the intensity,  especially the green channels.\n\n 2. hpa label can be classified as Enhanced, Supported, Approved, Uncertain? It is unclear that how kaggle data is labelled?\n\n\ndid anyone make some comparison of the followings:\n\n 1. duplicates are affecting public LB results (change of distribution) or not\n 2. effects of using kaggle only,  kaggle+external_hpa, external_hpa only,  kaggle+external_hpa, but finetune on hpa\n 3. if addition of external_hpa makes training/validation unstable\n\nThanks.\n\n \n\nhttps://www.proteinatlas.org/about/assays+annotation\n\n",
      "votes": 8
    },
    {
      "id": 448556,
      "postDate": "2019-01-01T13:40:57.077Z",
      "content": "<p>I have similar concerns but missed this thread and started something similar here: <a href=\"https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/76336\">https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/76336</a></p>",
      "rawMarkdown": "I have similar concerns but missed this thread and started something similar here: https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/76336"
    },
    {
      "id": 448395,
      "postDate": "2019-01-01T02:32:23.723Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 448556,
      "author_name": "Mark Worrall",
      "author_url": "",
      "post_date": "2019-01-01T13:40:57.077000",
      "content": "<p>I have similar concerns but missed this thread and started something similar here: <a href=\"https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/76336\">https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/76336</a></p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 448395,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-01-01T02:32:23.723000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "448225": "Is there a difference between the kaggle and the hpa external data?\n\n 1.  there seem to be a difference in the intensity,  especially the green channels.\n\n 2. hpa label can be classified as Enhanced, Supported, Approved, Uncertain? It is unclear that how kaggle data is labelled?\n\n\ndid anyone make some comparison of the followings:\n\n 1. duplicates are affecting public LB results (change of distribution) or not\n 2. effects of using kaggle only,  kaggle+external_hpa, external_hpa only,  kaggle+external_hpa, but finetune on hpa\n 3. if addition of external_hpa makes training/validation unstable\n\nThanks.\n\n \n\nhttps://www.proteinatlas.org/about/assays+annotation\n\n",
    "448556": "I have similar concerns but missed this thread and started something similar here: https://www.kaggle.com/c/human-protein-atlas-image-classification/discussion/76336",
    "448395": ""
  }
}