-
Notifications
You must be signed in to change notification settings - Fork 715
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Question about recall calculation #519
Comments
We look at the distances of the reported points. Every point that is within the distance (+ small epsilon) of the k-th nearest neighbor is considered a correct result. |
Suppose for k=10 an algorithm returns 10 copies of the same point which is the closest to a query. Will the recall be 100% or 10% in this case? |
Now I get what you mean. As far as I can see, we do not check whether an algorithm returns duplicated points. It would get 100% recall by just returning the nearest neighbor 10 times. |
OK, thanks. But now it seems this is a loophole for ANNS algorithms developers :). If someone fix it, the results may change a little. |
Hello,
Could you clarify please how recall is calculated in the case when an algorithm search returns 10 equal vectors each of which is the closest to the request? Is it possible that in such case the recall (from get_recall_values function?) will be 100%?
The text was updated successfully, but these errors were encountered: