You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
HierarchicalNSW will allow you to add two different vectors with the same label, which seems slightly surprising to me.
I'm not sure what to suggest here, since I don't know what the intended behavior is; I just wanted to bring it up. For my purposes, I might prefer the ability to overwrite the old value. I realize that may not suit everyone's needs, though. I presume it would depend on issue #4, too.
The text was updated successfully, but these errors were encountered:
Hi @SeanU, initially the labels where allowed to be duplicated (e.g. so that they can be used as classed for kNN classification), but after #57 is merged the labels will be required to be unique.
After the deletion will be added you would be able to overwrite old elements.
Hmm. That's a good point about making it more directly usable for KNN classification, though.
That said, if losing that feature is the cost of #57, maybe it's a net gain. It would be cheaper to maintain a map from labels to classes than it is to maintain a 2nd copy of all the vectors in memory for use cases like K-means clustering.
HierarchicalNSW
will allow you to add two different vectors with the same label, which seems slightly surprising to me.I'm not sure what to suggest here, since I don't know what the intended behavior is; I just wanted to bring it up. For my purposes, I might prefer the ability to overwrite the old value. I realize that may not suit everyone's needs, though. I presume it would depend on issue #4, too.
The text was updated successfully, but these errors were encountered: