{
  "id": 109119,
  "title": "Is HUVEC-05 mislabeled?",
  "url": "/competitions/recursion-cellular-image-classification/discussion/109119",
  "author_name": "",
  "post_date": "2019-09-16T16:38:27.307920200Z",
  "votes": 10,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Does anyone get a reasonable accuracy for HUVEC-05? If I have it in my validation experiments my score is always like random guessing. Is the experiment mislabeled or is there another explanation?</p>",
  "messages": [
    {
      "id": "627968",
      "postDate": "09/16/2019 16:38:27",
      "content": "<p>Does anyone get a reasonable accuracy for HUVEC-05? If I have it in my validation experiments my score is always like random guessing. Is the experiment mislabeled or is there another explanation?</p>",
      "rawMarkdown": "Does anyone get a reasonable accuracy for HUVEC-05? If I have it in my validation experiments my score is always like random guessing. Is the experiment mislabeled or is there another explanation?",
      "votes": null
    },
    {
      "id": "628002",
      "postDate": "09/16/2019 17:21:37",
      "content": "<p>Far away from random guessing, but I am getting validation accuracy 25% on it, while the next worst HUVEC experiment is 90%. I think it is some kind of between-experiments variation, but not clear what exactly. Thanks for posting, I thought it might be a bug in my code.</p>",
      "rawMarkdown": "Far away from random guessing, but I am getting validation accuracy 25% on it, while the next worst HUVEC experiment is 90%. I think it is some kind of between-experiments variation, but not clear what exactly. Thanks for posting, I thought it might be a bug in my code.",
      "votes": null
    },
    {
      "id": "629084",
      "postDate": "09/18/2019 10:05:04",
      "content": "<p>Ya it looks like a strong batch effect. A good way to see this is by looking at the average pixel value for each experiment\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F747148%2F4bcc8a74c51002d6fb22197ce551e17c%2FScreenshot%202019-09-18%20at%2010.54.07.png?generation=1568800596059485&amp;alt=media\" alt=\"\">\nthat big drop on the left is for HUVEV-05. You can also see there's big changes like that in the test set so it will be important to be able to control for those.</p>",
      "rawMarkdown": "Ya it looks like a strong batch effect. A good way to see this is by looking at the average pixel value for each experiment\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F747148%2F4bcc8a74c51002d6fb22197ce551e17c%2FScreenshot%202019-09-18%20at%2010.54.07.png?generation=1568800596059485&amp;alt=media)\nthat big drop on the left is for HUVEV-05. You can also see there's big changes like that in the test set so it will be important to be able to control for those.",
      "votes": null
    },
    {
      "id": "629448",
      "postDate": "09/18/2019 19:31:31",
      "content": "<p>Just look on image, its just the active substance penetrated into the cell much less</p>",
      "rawMarkdown": "Just look on image, its just the active substance penetrated into the cell much less",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 628002,
      "author_name": "zaharch",
      "author_url": "",
      "post_date": "09/16/2019 17:21:37",
      "content": "<p>Far away from random guessing, but I am getting validation accuracy 25% on it, while the next worst HUVEC experiment is 90%. I think it is some kind of between-experiments variation, but not clear what exactly. Thanks for posting, I thought it might be a bug in my code.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 629084,
      "author_name": "pflood314",
      "author_url": "",
      "post_date": "09/18/2019 10:05:04",
      "content": "<p>Ya it looks like a strong batch effect. A good way to see this is by looking at the average pixel value for each experiment\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F747148%2F4bcc8a74c51002d6fb22197ce551e17c%2FScreenshot%202019-09-18%20at%2010.54.07.png?generation=1568800596059485&amp;alt=media\" alt=\"\">\nthat big drop on the left is for HUVEV-05. You can also see there's big changes like that in the test set so it will be important to be able to control for those.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 629448,
      "author_name": "leighplt",
      "author_url": "",
      "post_date": "09/18/2019 19:31:31",
      "content": "<p>Just look on image, its just the active substance penetrated into the cell much less</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "627968": "Does anyone get a reasonable accuracy for HUVEC-05? If I have it in my validation experiments my score is always like random guessing. Is the experiment mislabeled or is there another explanation?",
    "628002": "Far away from random guessing, but I am getting validation accuracy 25% on it, while the next worst HUVEC experiment is 90%. I think it is some kind of between-experiments variation, but not clear what exactly. Thanks for posting, I thought it might be a bug in my code.",
    "629084": "Ya it looks like a strong batch effect. A good way to see this is by looking at the average pixel value for each experiment\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F747148%2F4bcc8a74c51002d6fb22197ce551e17c%2FScreenshot%202019-09-18%20at%2010.54.07.png?generation=1568800596059485&amp;alt=media)\nthat big drop on the left is for HUVEV-05. You can also see there's big changes like that in the test set so it will be important to be able to control for those.",
    "629448": "Just look on image, its just the active substance penetrated into the cell much less"
  },
  "source": "meta"
}