batch search queries

[arita37]

Currently,
one can query one vector a time the topk vector.

Do you think it is possible to have a batch search queries,
something like

search_batch(
vectors : list of numpy vectors.
fliter : filtering criteria, same for all input vectors.

return :

list of results :
res[0] : result for input vector index 0,
res[1] : result for input vector index 1,
res[2] : result for input vector index 2,

Technically, we can do it on client python side (multi threading),
but it might be more optimized to do it on server side (ie sharing the filtering criteria).

Standard KVS, RocksDB, Cassandra are doing this...

thanks for your good DB.

[generall]

Hi @arita37, it might be a useful feature, indeed! Thanks for your proposal

[arita37]

Hello,
Additionnally, we have this use case :

1 vector numpy.
mutiple filtering with OR operator:
filter_cond1 OR filter_cond2 OR filter_cond3

results :
filter_cond1 --> topk list
filter_cond2 --> topk list
filter_cond3 --> topk list

Typical use case is , when each vector is associated with category_id
vector, cat_id

We retrieve the top-k for cat_id in (2,5,7,9).

This is general use case and having the compute in batch inside the server
may be much faster... Mutiple queries for same vector takes extra time...