{
  "id": 166054,
  "title": "Huge gap between local validation score and LB",
  "url": "/competitions/global-wheat-detection/discussion/166054",
  "author_name": "Tu Nguyen",
  "post_date": "2020-07-11T17:45:26.916000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I've just joined this competition few days ago and I found that there is a big gap between local score and LB. When I used this <a href=\"https://www.kaggle.com/pestipeti/competition-metric-details-script\">script</a> (thanks Peter), I have 0.62 on local but 0.71 on LB. \nOtherwise, when I used pycocotools with a little modification by only calculating on the range 0.5-&gt;0.75 instead of 0.5-&gt;0.95 by default and adding the false negative in the precision formula, I had 0.76 on local and 0.61 on LB. </p>\n\n<p>Have anyone encountered the same behavior? What do you do on your validation? How do you split your data? anyone who did find a consistency between local score and LB could share some insights please?</p>\n\n<p>Thanks</p>",
  "messages": [
    {
      "id": 924917,
      "postDate": "2020-07-11T17:45:26.917Z",
      "content": "<p>I've just joined this competition few days ago and I found that there is a big gap between local score and LB. When I used this <a href=\"https://www.kaggle.com/pestipeti/competition-metric-details-script\">script</a> (thanks Peter), I have 0.62 on local but 0.71 on LB. \nOtherwise, when I used pycocotools with a little modification by only calculating on the range 0.5-&gt;0.75 instead of 0.5-&gt;0.95 by default and adding the false negative in the precision formula, I had 0.76 on local and 0.61 on LB. </p>\n\n<p>Have anyone encountered the same behavior? What do you do on your validation? How do you split your data? anyone who did find a consistency between local score and LB could share some insights please?</p>\n\n<p>Thanks</p>",
      "rawMarkdown": "I've just joined this competition few days ago and I found that there is a big gap between local score and LB. When I used this [script](https://www.kaggle.com/pestipeti/competition-metric-details-script) (thanks Peter), I have 0.62 on local but 0.71 on LB. \nOtherwise, when I used pycocotools with a little modification by only calculating on the range 0.5-&gt;0.75 instead of 0.5-&gt;0.95 by default and adding the false negative in the precision formula, I had 0.76 on local and 0.61 on LB. \n\nHave anyone encountered the same behavior? What do you do on your validation? How do you split your data? anyone who did find a consistency between local score and LB could share some insights please?\n\nThanks",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "924917": "I've just joined this competition few days ago and I found that there is a big gap between local score and LB. When I used this [script](https://www.kaggle.com/pestipeti/competition-metric-details-script) (thanks Peter), I have 0.62 on local but 0.71 on LB. \nOtherwise, when I used pycocotools with a little modification by only calculating on the range 0.5-&gt;0.75 instead of 0.5-&gt;0.95 by default and adding the false negative in the precision formula, I had 0.76 on local and 0.61 on LB. \n\nHave anyone encountered the same behavior? What do you do on your validation? How do you split your data? anyone who did find a consistency between local score and LB could share some insights please?\n\nThanks"
  }
}