{
  "id": 175320,
  "title": "Such a tough Competition. Was it really CV?",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/175320",
  "author_name": "Indranil Bhattacharya",
  "post_date": "2020-08-18T00:18:18.986000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Our team trusted on CV and it seems like submission with lower cv had higher private LB score. We used Chris's triple stratified CV strategy with the 2020 set only. We can see how close the LB scores are, to reduce the variance we had built 20 models (eff nets, dense-nets, res-nexts, lightgbm on ICA and PCA features with the features extracted from last layers etc. ). But with no luck.</p>\n<p>Looking forward to seeing the winner's solution and learn what we missed. GGWP everyone.</p>",
  "messages": [
    {
      "id": 974418,
      "postDate": "2020-08-18T00:18:18.987Z",
      "content": "<p>Our team trusted on CV and it seems like submission with lower cv had higher private LB score. We used Chris's triple stratified CV strategy with the 2020 set only. We can see how close the LB scores are, to reduce the variance we had built 20 models (eff nets, dense-nets, res-nexts, lightgbm on ICA and PCA features with the features extracted from last layers etc. ). But with no luck.</p>\n<p>Looking forward to seeing the winner's solution and learn what we missed. GGWP everyone.</p>",
      "rawMarkdown": "Our team trusted on CV and it seems like submission with lower cv had higher private LB score. We used Chris's triple stratified CV strategy with the 2020 set only. We can see how close the LB scores are, to reduce the variance we had built 20 models (eff nets, dense-nets, res-nexts, lightgbm on ICA and PCA features with the features extracted from last layers etc. ). But with no luck.\n\nLooking forward to seeing the winner's solution and learn what we missed. GGWP everyone.",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "974418": "Our team trusted on CV and it seems like submission with lower cv had higher private LB score. We used Chris's triple stratified CV strategy with the 2020 set only. We can see how close the LB scores are, to reduce the variance we had built 20 models (eff nets, dense-nets, res-nexts, lightgbm on ICA and PCA features with the features extracted from last layers etc. ). But with no luck.\n\nLooking forward to seeing the winner's solution and learn what we missed. GGWP everyone."
  }
}