Hi, are there any plans to include "vector search"...
# community-helps
Stefan Hesse
10/19/2021, 7:40 AMHi, are there any plans to include "vector search" aka search by cosine similarity or euclidean distance?
k
Kishore Nallan
10/19/2021, 7:40 AMYes, that's on our radar.
Kishore Nallan
10/19/2021, 7:41 AMCan you please elaborate on your use case? Do you need that to work along with the search component?
s
Stefan Hesse
10/19/2021, 7:43 AMRight now I want to use it like this:
I pre-calculate embeddings for product name and would then like to use typesense to find duplicates.
So wold be mostly used for backend
Stefan Hesse
10/19/2021, 7:44 AMpostgresql's cube do not cut it. I think elastic-search has something like this
k
Kishore Nallan
10/19/2021, 7:45 AMGot it. Yeah we want to support a
vector[X]s
Stefan Hesse
10/19/2021, 7:46 AMthat should hit my use case, assuming I can choose the function for nearest neighbor
k
Kishore Nallan
10/19/2021, 7:47 AMWill likely start with cosine I think. Will depend on the library used and data set size (approximation places some restrictions on the metric used).
s
Stefan Hesse
10/19/2021, 7:48 AMcosine is what I need, so perfect
k
Kishore Nallan
10/19/2021, 7:48 AMHow are you calculating the embedding? Word2vec / fasttext?
s
Stefan Hesse
10/19/2021, 7:49 AMI am using sbert.net
k
Kishore Nallan
10/19/2021, 7:49 AM👍