{
  "id": 456342,
  "title": "TMAs in test set might be quite different from those in the training set?",
  "url": "/competitions/UBC-OCEAN/discussion/456342",
  "author_name": "Sihan Lv",
  "post_date": "2023-11-19T15:44:22.692000",
  "votes": 4,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I trained models with clipped images from WSIs and got a balanced accuary of 0.8 on TMAs in the training set, but it didn't improve my LB score.</p>\n<p>That's strange. Have you guys had anything like that?</p>",
  "messages": [
    {
      "id": 2530874,
      "postDate": "2023-11-19T15:44:22.693Z",
      "content": "<p>I trained models with clipped images from WSIs and got a balanced accuary of 0.8 on TMAs in the training set, but it didn't improve my LB score.</p>\n<p>That's strange. Have you guys had anything like that?</p>",
      "rawMarkdown": "I trained models with clipped images from WSIs and got a balanced accuary of 0.8 on TMAs in the training set, but it didn't improve my LB score.\n\nThat's strange. Have you guys had anything like that?",
      "votes": 4
    },
    {
      "id": 2530878,
      "postDate": "2023-11-19T15:47:12.083Z",
      "content": "<p>Note: I used TMAs from the training set as a validation set and did not use them for training</p>",
      "rawMarkdown": "Note: I used TMAs from the training set as a validation set and did not use them for training",
      "replies": [
        {
          "id": 2530880,
          "postDate": "2023-11-19T15:51:01.960Z",
          "content": "<p>me too… 😂</p>",
          "rawMarkdown": "me too... 😂",
          "replies": [
            {
              "id": 2530884,
              "postDate": "2023-11-19T15:58:33.973Z",
              "content": "<p>Well, it looks like it's our job to solve this.🤣</p>",
              "rawMarkdown": "Well, it looks like it's our job to solve this.🤣"
            }
          ]
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 2530878,
      "author_name": "Sihan Lv",
      "author_url": "",
      "post_date": "2023-11-19T15:47:12.083000",
      "content": "<p>Note: I used TMAs from the training set as a validation set and did not use them for training</p>",
      "votes": 0,
      "replies": [
        {
          "id": 2530880,
          "author_name": "Quan Vu",
          "author_url": "",
          "post_date": "2023-11-19T15:51:01.960000",
          "content": "<p>me too… 😂</p>",
          "votes": 0,
          "replies": [
            {
              "id": 2530884,
              "author_name": "Sihan Lv",
              "author_url": "",
              "post_date": "2023-11-19T15:58:33.973000",
              "content": "<p>Well, it looks like it's our job to solve this.🤣</p>",
              "votes": 0,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2530874": "I trained models with clipped images from WSIs and got a balanced accuary of 0.8 on TMAs in the training set, but it didn't improve my LB score.\n\nThat's strange. Have you guys had anything like that?",
    "2530878": "Note: I used TMAs from the training set as a validation set and did not use them for training"
  }
}