Skip to main content
29 events
when toggle format what by license comment
Apr 12, 2019 at 12:17 vote accept tkarahan
Apr 12, 2019 at 12:16 comment added tkarahan Thank you so much. It really helped to clarify subject.
Apr 12, 2019 at 12:15 comment added Esmailian @TolgaKarahan Exactly.
Apr 12, 2019 at 12:12 history edited Esmailian CC BY-SA 4.0
Explanation improved
Apr 12, 2019 at 12:09 comment added tkarahan It's good. Finally If I define TP / P as positive recall and TN / N as negative recall then I suppose with increasing precision I increase negative recall and with increasing recall because it is same thing I also increase positive recall. So it looks like matter of increasing negative or positive recall and which one more important to me.
Apr 12, 2019 at 11:59 comment added Esmailian @TolgaKarahan Aha. For better-than-random models, increase in precision means decrease in recall (and vice versa), which is decrease in TP/P (P = TP+FN). For TN/N, we know when threshold is increased (decrease in recall) both TP and FP decrease since we are selecting less positives, thus FP/N decreases, and 1 - FP/N = TN/N increases. So the answer to your question is yes.
Apr 12, 2019 at 11:47 comment added tkarahan I mean ratio of correctly predicted labels to all labels for a specific class. Like TP / P or TN / N. If I increase precision does it predict negative examples more accurately with increasing TN / N?
Apr 12, 2019 at 11:24 history edited Esmailian CC BY-SA 4.0
Caption corrected
Apr 12, 2019 at 11:15 comment added Esmailian @TolgaKarahan First we need to define "more accurately" in terms of TN, TP, etc. For example "accuracy" is for both positives and negatives, i.e. (TP+TN / P+N) which I added it to the plots, it has a rise and a fall for better-than-random models.
Apr 12, 2019 at 11:12 history edited Esmailian CC BY-SA 4.0
Explanation improved
Apr 12, 2019 at 8:13 comment added tkarahan So when random phenomena completely rules, in practice it is observed that they generally have inverse relationship. There are different situations but, can we say generally if we increase precision it means that we predict negative examples more accurately and if we increase recall it means that we predict positive examples more accurately?
Apr 12, 2019 at 1:27 history edited Esmailian CC BY-SA 4.0
Explanation improved
Apr 12, 2019 at 0:59 history edited Esmailian CC BY-SA 4.0
Explanation improved
Apr 12, 2019 at 0:33 history edited Esmailian CC BY-SA 4.0
Explanation improved
Apr 12, 2019 at 0:02 history edited Esmailian CC BY-SA 4.0
Code corrected
Apr 11, 2019 at 23:56 history undeleted Esmailian
Apr 11, 2019 at 23:56 history edited Esmailian CC BY-SA 4.0
Explanation improved
Apr 11, 2019 at 22:34 history deleted Esmailian via Vote
Apr 11, 2019 at 19:47 history edited Esmailian CC BY-SA 4.0
Explanation improved
Apr 11, 2019 at 19:32 history edited Esmailian CC BY-SA 4.0
Explanation improved
Apr 11, 2019 at 19:18 history edited Esmailian CC BY-SA 4.0
added 34 characters in body
Apr 11, 2019 at 19:11 history edited Esmailian CC BY-SA 4.0
Explanation improved
Apr 11, 2019 at 18:47 history edited Esmailian CC BY-SA 4.0
Explanation improved
Apr 11, 2019 at 18:14 history edited Esmailian CC BY-SA 4.0
Explanation improved
Apr 11, 2019 at 18:07 history edited Esmailian CC BY-SA 4.0
Explanation improved
Apr 11, 2019 at 18:02 history undeleted Esmailian
Apr 11, 2019 at 18:01 history deleted Esmailian via Vote
Apr 11, 2019 at 17:59 history edited Esmailian CC BY-SA 4.0
added 1 character in body
Apr 11, 2019 at 17:53 history answered Esmailian CC BY-SA 4.0