{
  "id": 236569,
  "title": "How to calculate FN?",
  "url": "/competitions/coleridgeinitiative-show-us-the-data/discussion/236569",
  "author_name": "Xiaonan Ji",
  "post_date": "2021-05-05T01:07:02.275000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>In the <a href=\"https://www.kaggle.com/c/coleridgeinitiative-show-us-the-data/overview/evaluation\" target=\"_blank\">evaluation page</a>, it says: \"Any ground truths with no nearest predictions are counted as false negatives (FN)\". I don't quite understand what is \"no nearest predictions\"? Does it mean all candidates' jaccoard scores are 0.0?</p>\n<p>[Edit] Also another question: if a singal prediction is best matched to one actual label, it will be treated as TP. After this, could the same prediction be best matched to a second actual label and remained TP? Or for the second actual we have to pick another prediction that hasn't been treated as TP? In other words, can a single prediction be the best matches to multiple actual labels and be \"multi-tagged\" as TP (of course when calculating precisions and recalls we will count it only once)?</p>\n<p>Thank you for the help.</p>",
  "messages": [
    {
      "id": 1293517,
      "postDate": "2021-05-05T01:07:02.277Z",
      "content": "<p>In the <a href=\"https://www.kaggle.com/c/coleridgeinitiative-show-us-the-data/overview/evaluation\" target=\"_blank\">evaluation page</a>, it says: \"Any ground truths with no nearest predictions are counted as false negatives (FN)\". I don't quite understand what is \"no nearest predictions\"? Does it mean all candidates' jaccoard scores are 0.0?</p>\n<p>[Edit] Also another question: if a singal prediction is best matched to one actual label, it will be treated as TP. After this, could the same prediction be best matched to a second actual label and remained TP? Or for the second actual we have to pick another prediction that hasn't been treated as TP? In other words, can a single prediction be the best matches to multiple actual labels and be \"multi-tagged\" as TP (of course when calculating precisions and recalls we will count it only once)?</p>\n<p>Thank you for the help.</p>",
      "rawMarkdown": "In the [evaluation page](https://www.kaggle.com/c/coleridgeinitiative-show-us-the-data/overview/evaluation), it says: \"Any ground truths with no nearest predictions are counted as false negatives (FN)\". I don't quite understand what is \"no nearest predictions\"? Does it mean all candidates' jaccoard scores are 0.0?\n\n[Edit] Also another question: if a singal prediction is best matched to one actual label, it will be treated as TP. After this, could the same prediction be best matched to a second actual label and remained TP? Or for the second actual we have to pick another prediction that hasn't been treated as TP? In other words, can a single prediction be the best matches to multiple actual labels and be \"multi-tagged\" as TP (of course when calculating precisions and recalls we will count it only once)?\n\nThank you for the help.",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1293517": "In the [evaluation page](https://www.kaggle.com/c/coleridgeinitiative-show-us-the-data/overview/evaluation), it says: \"Any ground truths with no nearest predictions are counted as false negatives (FN)\". I don't quite understand what is \"no nearest predictions\"? Does it mean all candidates' jaccoard scores are 0.0?\n\n[Edit] Also another question: if a singal prediction is best matched to one actual label, it will be treated as TP. After this, could the same prediction be best matched to a second actual label and remained TP? Or for the second actual we have to pick another prediction that hasn't been treated as TP? In other words, can a single prediction be the best matches to multiple actual labels and be \"multi-tagged\" as TP (of course when calculating precisions and recalls we will count it only once)?\n\nThank you for the help."
  }
}