With vector search algorithms and databases, it is common to tune the configurations of the index that govern how it is built as well as tune some parameters during search time. For example, with HNSW, one can tune maxConnections or efConstruction prior to building the index and efSearch during querying. How can I get such control over the index configuration within Pinecone? I know there is a index_config parameter that can be passed to pinecone.create_index(), but it is poorly documented and I figure out how to set up my desired configuration. If someone could share an example with the HNSW config parameters listed above, that would be greatly appreciated!!
Hi @vgwinnr!
We actually don’t use HNSW to power any of our indexes (keep an eye on our blog for a writeup on just why that is in the next day or two).
Since Pinecone is a managed service, our goal is to remove the need for developers to spend their time endlessly tuning complex algorithms. So, we only expose a few configuration levers.
What type of configuration are you having trouble making?
The main levers we expose are pod type, similarity metric, dimensions, pod sizes, and the # pods (replicas).
LMK!