{
  "id": 386658,
  "title": "Do test-time augmentation or ensemble methods improve model performance?",
  "url": "/competitions/rsna-breast-cancer-detection/discussion/386658",
  "author_name": "ECO",
  "post_date": "2023-02-14T05:05:58.561000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I understand that many Kagglers use test-time augmentation or multi-model ensembles to boost performance on the test set. </p>\n<p>However, despite trying cross-validation ensembles and horizontal/vertical flip test-time augmentations, I have not seen any improvement in my LB score. </p>\n<p>In fact, the multi-prediction scores are lower than those of a single CNN (LB 0.57). </p>\n<p>Do you have any suggestions?</p>\n<p>Thank you.</p>",
  "messages": [
    {
      "id": 2143154,
      "postDate": "2023-02-14T05:05:58.560Z",
      "content": "<p>I understand that many Kagglers use test-time augmentation or multi-model ensembles to boost performance on the test set. </p>\n<p>However, despite trying cross-validation ensembles and horizontal/vertical flip test-time augmentations, I have not seen any improvement in my LB score. </p>\n<p>In fact, the multi-prediction scores are lower than those of a single CNN (LB 0.57). </p>\n<p>Do you have any suggestions?</p>\n<p>Thank you.</p>",
      "rawMarkdown": "I understand that many Kagglers use test-time augmentation or multi-model ensembles to boost performance on the test set. \n\nHowever, despite trying cross-validation ensembles and horizontal/vertical flip test-time augmentations, I have not seen any improvement in my LB score. \n\nIn fact, the multi-prediction scores are lower than those of a single CNN (LB 0.57). \n\nDo you have any suggestions?\n\nThank you.",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2143154": "I understand that many Kagglers use test-time augmentation or multi-model ensembles to boost performance on the test set. \n\nHowever, despite trying cross-validation ensembles and horizontal/vertical flip test-time augmentations, I have not seen any improvement in my LB score. \n\nIn fact, the multi-prediction scores are lower than those of a single CNN (LB 0.57). \n\nDo you have any suggestions?\n\nThank you."
  }
}