{
  "id": 125947,
  "title": "Why big diff between Local CV score and LB",
  "url": "/competitions/pku-autonomous-driving/discussion/125947",
  "author_name": "",
  "post_date": "2020-01-14T17:32:41.138624100Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi\nCould any one be able to explain the big difference between CV score and LB score. I use the same script that was published for calculating the mAP . \nI can touch mAP as high as .198 with reasonable balance between Val loss and Train loss but fail to even hit .08 with this on leader board. </p>",
  "messages": [
    {
      "id": "718725",
      "postDate": "01/14/2020 17:32:41",
      "content": "<p>Hi\nCould any one be able to explain the big difference between CV score and LB score. I use the same script that was published for calculating the mAP . \nI can touch mAP as high as .198 with reasonable balance between Val loss and Train loss but fail to even hit .08 with this on leader board. </p>",
      "rawMarkdown": "Hi\nCould any one be able to explain the big difference between CV score and LB score. I use the same script that was published for calculating the mAP . \nI can touch mAP as high as .198 with reasonable balance between Val loss and Train loss but fail to even hit .08 with this on leader board.",
      "votes": null
    },
    {
      "id": "718831",
      "postDate": "01/14/2020 20:59:33",
      "content": "<p>I suspect it doesn't count for false positives. Predicts that haven't matched with any gt are left in the dust.. Maybe you can add some penalty for these ones</p>",
      "rawMarkdown": "I suspect it doesn't count for false positives. Predicts that haven't matched with any gt are left in the dust.. Maybe you can add some penalty for these ones",
      "votes": null
    },
    {
      "id": "718914",
      "postDate": "01/15/2020 00:29:47",
      "content": "<p>I had a strange finding. My base model gave lb 0.07 with threshold 0.3, while I raise the threshold to 0.535, lb score was still 0.07.</p>",
      "rawMarkdown": "I had a strange finding. My base model gave lb 0.07 with threshold 0.3, while I raise the threshold to 0.535, lb score was still 0.07.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 718831,
      "author_name": "dvorobiev",
      "author_url": "",
      "post_date": "01/14/2020 20:59:33",
      "content": "<p>I suspect it doesn't count for false positives. Predicts that haven't matched with any gt are left in the dust.. Maybe you can add some penalty for these ones</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 718914,
      "author_name": "richardwong1994",
      "author_url": "",
      "post_date": "01/15/2020 00:29:47",
      "content": "<p>I had a strange finding. My base model gave lb 0.07 with threshold 0.3, while I raise the threshold to 0.535, lb score was still 0.07.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "718725": "Hi\nCould any one be able to explain the big difference between CV score and LB score. I use the same script that was published for calculating the mAP . \nI can touch mAP as high as .198 with reasonable balance between Val loss and Train loss but fail to even hit .08 with this on leader board.",
    "718831": "I suspect it doesn't count for false positives. Predicts that haven't matched with any gt are left in the dust.. Maybe you can add some penalty for these ones",
    "718914": "I had a strange finding. My base model gave lb 0.07 with threshold 0.3, while I raise the threshold to 0.535, lb score was still 0.07."
  },
  "source": "meta"
}