Apparently \"delete\" is not an option. :(
", "messages": [ { "id": "204235", "postDate": "07/17/2017 23:52:05", "content": "Apparently \"delete\" is not an option. :(
", "rawMarkdown": "Apparently \"delete\" is not an option. :(", "votes": null }, { "id": "204408", "postDate": "07/18/2017 12:52:17", "content": "\n\n\nThe better criteria for this challenge is to detect all of the known threats with the least number of false positive detections.
\n
Calibrated probabilities (the kind which log loss rewards) allow you to choose your false positive / false negative thresholds however you please. The competition is not being judged on hard classifications.
", "rawMarkdown": "> The better criteria for this challenge is to detect all of the known threats with the least number of false positive detections.\n\nCalibrated probabilities (the kind which log loss rewards) allow you to choose your false positive / false negative thresholds however you please. The competition is not being judged on hard classifications.", "votes": null }, { "id": "207122", "postDate": "07/25/2017 19:26:24", "content": "One would think that a \"better criteria\" would be to minimize the false negatives. A false positive, is a hassle, have to check person only to find nothing; but it's a safer situation. Whereas, a false negative means someone got on plane with something they likely should not have. Unless I'm backwards on the criteria, this seems like the thing to do.
", "rawMarkdown": "One would think that a \"better criteria\" would be to minimize the false negatives. A false positive, is a hassle, have to check person only to find nothing; but it's a safer situation. Whereas, a false negative means someone got on plane with something they likely should not have. Unless I'm backwards on the criteria, this seems like the thing to do.", "votes": null }, { "id": "207468", "postDate": "07/26/2017 18:01:37", "content": "Minimizing just false negatives can be gamed by guessing positive for every example. Like William said, predicting a threshold should let the person choose the trade-off.
", "rawMarkdown": "Minimizing just false negatives can be gamed by guessing positive for every example. Like William said, predicting a threshold should let the person choose the trade-off.", "votes": null }, { "id": "207516", "postDate": "07/26/2017 19:42:43", "content": "Likewise can minimize false positives by guessing all negatives. But, my point, which should've been stated better, was that threshold should be chosen to minimize false negatives. I'm basing this totally on idea that false negatives are more dangerous than false positives; but I'm open to other perspectives.
", "rawMarkdown": "Likewise can minimize false positives by guessing all negatives. But, my point, which should've been stated better, was that threshold should be chosen to minimize false negatives. I'm basing this totally on idea that false negatives are more dangerous than false positives; but I'm open to other perspectives.", "votes": null } ], "comments": [ { "id": 204408, "author_name": "wcukierski", "author_url": "", "post_date": "07/18/2017 12:52:17", "content": "\n\n\nThe better criteria for this challenge is to detect all of the known threats with the least number of false positive detections.
\n
Calibrated probabilities (the kind which log loss rewards) allow you to choose your false positive / false negative thresholds however you please. The competition is not being judged on hard classifications.
", "votes": null, "replies": [] }, { "id": 207122, "author_name": "srdhaeayautnon", "author_url": "", "post_date": "07/25/2017 19:26:24", "content": "One would think that a \"better criteria\" would be to minimize the false negatives. A false positive, is a hassle, have to check person only to find nothing; but it's a safer situation. Whereas, a false negative means someone got on plane with something they likely should not have. Unless I'm backwards on the criteria, this seems like the thing to do.
", "votes": null, "replies": [ { "id": 207468, "author_name": "act444", "author_url": "", "post_date": "07/26/2017 18:01:37", "content": "Minimizing just false negatives can be gamed by guessing positive for every example. Like William said, predicting a threshold should let the person choose the trade-off.
", "votes": null, "replies": [] } ] }, { "id": 207516, "author_name": "srdhaeayautnon", "author_url": "", "post_date": "07/26/2017 19:42:43", "content": "Likewise can minimize false positives by guessing all negatives. But, my point, which should've been stated better, was that threshold should be chosen to minimize false negatives. I'm basing this totally on idea that false negatives are more dangerous than false positives; but I'm open to other perspectives.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "204235": "Apparently \"delete\" is not an option. :(", "204408": "> The better criteria for this challenge is to detect all of the known threats with the least number of false positive detections.\n\nCalibrated probabilities (the kind which log loss rewards) allow you to choose your false positive / false negative thresholds however you please. The competition is not being judged on hard classifications.", "207122": "One would think that a \"better criteria\" would be to minimize the false negatives. A false positive, is a hassle, have to check person only to find nothing; but it's a safer situation. Whereas, a false negative means someone got on plane with something they likely should not have. Unless I'm backwards on the criteria, this seems like the thing to do.", "207468": "Minimizing just false negatives can be gamed by guessing positive for every example. Like William said, predicting a threshold should let the person choose the trade-off.", "207516": "Likewise can minimize false positives by guessing all negatives. But, my point, which should've been stated better, was that threshold should be chosen to minimize false negatives. I'm basing this totally on idea that false negatives are more dangerous than false positives; but I'm open to other perspectives." }, "source": "meta" }