{
  "id": 240759,
  "title": "study string format",
  "url": "/competitions/siim-covid19-detection/discussion/240759",
  "author_name": "",
  "post_date": "2021-05-21T09:14:43.789560900Z",
  "votes": 7,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Because the calculation method is MAP. So the best format is: <code>negative {conf} 0 0 1 1 typical {conf} 0 0 1 1 indeterminate {conf} 0 0 1 1 atypical {conf} 0 0 1 1</code>. Don't worry about false positives. It is very similar to the concept of AUC. The confidence value sorting is the key point.</p>\n<p>0.05 to 0.176<br>\nJust change the <code>negative 1 0 0 1 1</code> to <code>negative 1 0 0 1 1 typical 1 0 0 1 1 indeterminate 1 0 0 1 1 atypical 1 0 0 1 1</code> </p>\n<p>There should be a problem with the current image level. I also specially trained a 2-class model for (none). But the score is also only 0.001.</p>",
  "messages": [
    {
      "id": "1317277",
      "postDate": "05/21/2021 09:14:43",
      "content": "<p>Because the calculation method is MAP. So the best format is: <code>negative {conf} 0 0 1 1 typical {conf} 0 0 1 1 indeterminate {conf} 0 0 1 1 atypical {conf} 0 0 1 1</code>. Don't worry about false positives. It is very similar to the concept of AUC. The confidence value sorting is the key point.</p>\n<p>0.05 to 0.176<br>\nJust change the <code>negative 1 0 0 1 1</code> to <code>negative 1 0 0 1 1 typical 1 0 0 1 1 indeterminate 1 0 0 1 1 atypical 1 0 0 1 1</code> </p>\n<p>There should be a problem with the current image level. I also specially trained a 2-class model for (none). But the score is also only 0.001.</p>",
      "rawMarkdown": "Because the calculation method is MAP. So the best format is: `negative {conf} 0 0 1 1 typical {conf} 0 0 1 1 indeterminate {conf} 0 0 1 1 atypical {conf} 0 0 1 1`. Don't worry about false positives. It is very similar to the concept of AUC. The confidence value sorting is the key point.\n\n0.05 to 0.176\nJust change the `negative 1 0 0 1 1` to `negative 1 0 0 1 1 typical 1 0 0 1 1 indeterminate 1 0 0 1 1 atypical 1 0 0 1 1` \n\nThere should be a problem with the current image level. I also specially trained a 2-class model for (none). But the score is also only 0.001.",
      "votes": null
    },
    {
      "id": "1317423",
      "postDate": "05/21/2021 12:01:51",
      "content": "<p>0.176 owing to sorted at random :)<br>\nMy understanding is that the full score of study level classification and image level detection is 0.666 and 0.333, respectively.<br>\nFor image level, currently I encounter the same issue. <br>\nSurprisingly, format change from pascal to coco of submission.csv changed the LB score by only 0.001.<br>\nAny error or problem in scoring? Or just my mistake? <a href=\"https://www.kaggle.com/juliaelliott\" target=\"_blank\">@juliaelliott</a> <a href=\"https://www.kaggle.com/paras42\" target=\"_blank\">@paras42</a></p>",
      "rawMarkdown": "0.176 owing to sorted at random :)\nMy understanding is that the full score of study level classification and image level detection is 0.666 and 0.333, respectively.\nFor image level, currently I encounter the same issue. \nSurprisingly, format change from pascal to coco of submission.csv changed the LB score by only 0.001.\nAny error or problem in scoring? Or just my mistake? @juliaelliott @paras42",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1317423,
      "author_name": "drtausamaru",
      "author_url": "",
      "post_date": "05/21/2021 12:01:51",
      "content": "<p>0.176 owing to sorted at random :)<br>\nMy understanding is that the full score of study level classification and image level detection is 0.666 and 0.333, respectively.<br>\nFor image level, currently I encounter the same issue. <br>\nSurprisingly, format change from pascal to coco of submission.csv changed the LB score by only 0.001.<br>\nAny error or problem in scoring? Or just my mistake? <a href=\"https://www.kaggle.com/juliaelliott\" target=\"_blank\">@juliaelliott</a> <a href=\"https://www.kaggle.com/paras42\" target=\"_blank\">@paras42</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1317277": "Because the calculation method is MAP. So the best format is: `negative {conf} 0 0 1 1 typical {conf} 0 0 1 1 indeterminate {conf} 0 0 1 1 atypical {conf} 0 0 1 1`. Don't worry about false positives. It is very similar to the concept of AUC. The confidence value sorting is the key point.\n\n0.05 to 0.176\nJust change the `negative 1 0 0 1 1` to `negative 1 0 0 1 1 typical 1 0 0 1 1 indeterminate 1 0 0 1 1 atypical 1 0 0 1 1` \n\nThere should be a problem with the current image level. I also specially trained a 2-class model for (none). But the score is also only 0.001.",
    "1317423": "0.176 owing to sorted at random :)\nMy understanding is that the full score of study level classification and image level detection is 0.666 and 0.333, respectively.\nFor image level, currently I encounter the same issue. \nSurprisingly, format change from pascal to coco of submission.csv changed the LB score by only 0.001.\nAny error or problem in scoring? Or just my mistake? @juliaelliott @paras42"
  },
  "source": "meta"
}