{
  "id": 271945,
  "title": "How will experts choose the two final submission ?",
  "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/271945",
  "author_name": "Shummi Ahmed",
  "post_date": "2021-09-13T10:35:26.721000",
  "votes": 0,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Since competition is going towards end, i guess its right time to ask this. </p>\n<p>The public test set is only 22% of total test set, what things make you select between your entries which are different techniques and have almost same scores, for rest 78 % of data.</p>\n<p>Since on topic, what are <strong>training</strong> roc leading LB teams are getting?  (i am getting almost 95% roc)</p>\n<p>My strategy would be </p>\n<ol>\n<li>See best public lb result ofc</li>\n<li>Compute data correlation and variability within training data and other publicly available datasets to see the margins of variability in BT Scans, hopefully this will increase confidence that 22% of test is best representative of rest 78%. </li>\n<li>Training data </li>\n</ol>\n<p>As per my current hypothesis the 22% is not the best representative of training set as per my insignificant analysis :( </p>",
  "messages": [
    {
      "id": 1511328,
      "postDate": "2021-09-13T10:35:26.720Z",
      "content": "<p>Since competition is going towards end, i guess its right time to ask this. </p>\n<p>The public test set is only 22% of total test set, what things make you select between your entries which are different techniques and have almost same scores, for rest 78 % of data.</p>\n<p>Since on topic, what are <strong>training</strong> roc leading LB teams are getting?  (i am getting almost 95% roc)</p>\n<p>My strategy would be </p>\n<ol>\n<li>See best public lb result ofc</li>\n<li>Compute data correlation and variability within training data and other publicly available datasets to see the margins of variability in BT Scans, hopefully this will increase confidence that 22% of test is best representative of rest 78%. </li>\n<li>Training data </li>\n</ol>\n<p>As per my current hypothesis the 22% is not the best representative of training set as per my insignificant analysis :( </p>",
      "rawMarkdown": "Since competition is going towards end, i guess its right time to ask this. \n\nThe public test set is only 22% of total test set, what things make you select between your entries which are different techniques and have almost same scores, for rest 78 % of data.\n\nSince on topic, what are **training** roc leading LB teams are getting?  (i am getting almost 95% roc)\n\nMy strategy would be \n1. See best public lb result ofc\n2. Compute data correlation and variability within training data and other publicly available datasets to see the margins of variability in BT Scans, hopefully this will increase confidence that 22% of test is best representative of rest 78%. \n3. Training data \n\nAs per my current hypothesis the 22% is not the best representative of training set as per my insignificant analysis :( "
    },
    {
      "id": 1514003,
      "postDate": "2021-09-15T15:38:28.830Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1514003,
      "author_name": "",
      "author_url": "",
      "post_date": "2021-09-15T15:38:28.830000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1511328": "Since competition is going towards end, i guess its right time to ask this. \n\nThe public test set is only 22% of total test set, what things make you select between your entries which are different techniques and have almost same scores, for rest 78 % of data.\n\nSince on topic, what are **training** roc leading LB teams are getting?  (i am getting almost 95% roc)\n\nMy strategy would be \n1. See best public lb result ofc\n2. Compute data correlation and variability within training data and other publicly available datasets to see the margins of variability in BT Scans, hopefully this will increase confidence that 22% of test is best representative of rest 78%. \n3. Training data \n\nAs per my current hypothesis the 22% is not the best representative of training set as per my insignificant analysis :( ",
    "1514003": ""
  }
}