{
  "id": 108081,
  "title": "what's your highest private score? why you did not choose that one?",
  "url": "/competitions/aptos2019-blindness-detection/discussion/108081",
  "author_name": "",
  "post_date": "2019-09-09T00:09:13.376910Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>My highest private score is 0.928. \nThe reason I choose the current two 0.923 0.924 is trying to change the weights of ensemble and round threshold to get higher in private rather than highest in public. \nBecause this is no correlation between training data, private and public test, it is not easy to choose the correct one.</p>",
  "messages": [
    {
      "id": "621781",
      "postDate": "09/09/2019 00:09:13",
      "content": "<p>My highest private score is 0.928. \nThe reason I choose the current two 0.923 0.924 is trying to change the weights of ensemble and round threshold to get higher in private rather than highest in public. \nBecause this is no correlation between training data, private and public test, it is not easy to choose the correct one.</p>",
      "rawMarkdown": "My highest private score is 0.928. \nThe reason I choose the current two 0.923 0.924 is trying to change the weights of ensemble and round threshold to get higher in private rather than highest in public. \nBecause this is no correlation between training data, private and public test, it is not easy to choose the correct one.",
      "votes": null
    },
    {
      "id": "622750",
      "postDate": "09/10/2019 03:29:53",
      "content": "<p>My highest is 0.923 but it was a simply trained model that I didn't expect to do well. I wrote about it here: <a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108143\">https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108143</a>.</p>\n\n<p>I ended up choosing some stacked efficient nets that I trained differently but it didn't perform well at all on the private. I thought it would be the most robust. </p>",
      "rawMarkdown": "My highest is 0.923 but it was a simply trained model that I didn't expect to do well. I wrote about it here: https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108143.\n\nI ended up choosing some stacked efficient nets that I trained differently but it didn't perform well at all on the private. I thought it would be the most robust.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 622750,
      "author_name": "sterls",
      "author_url": "",
      "post_date": "09/10/2019 03:29:53",
      "content": "<p>My highest is 0.923 but it was a simply trained model that I didn't expect to do well. I wrote about it here: <a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108143\">https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108143</a>.</p>\n\n<p>I ended up choosing some stacked efficient nets that I trained differently but it didn't perform well at all on the private. I thought it would be the most robust. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "621781": "My highest private score is 0.928. \nThe reason I choose the current two 0.923 0.924 is trying to change the weights of ensemble and round threshold to get higher in private rather than highest in public. \nBecause this is no correlation between training data, private and public test, it is not easy to choose the correct one.",
    "622750": "My highest is 0.923 but it was a simply trained model that I didn't expect to do well. I wrote about it here: https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/108143.\n\nI ended up choosing some stacked efficient nets that I trained differently but it didn't perform well at all on the private. I thought it would be the most robust."
  },
  "source": "meta"
}