CODEWITHSUNDEEP
indexingcluster-analysisnearest-neighborannoy
Chaitanya Patil
Chaitanya Patil

Reputation: 333

How does Annoy Index the embeddings?

I am trying to understand how Annoy Indexing works..I have referred the following documents: https://github.com/spotify/annoy#how-does-it-work
https://cloud.google.com/solutions/machine-learning/building-real-time-embeddings-similarity-matching-system
These documents explain how to get index from annoy but it does not explain HOW the Indexes are created?

Lets say I have sentence embedding matrix of 3 dimension (for simplicity)

[[1,2,3]                
[4,2,3]             
[1,2,3]             
[1,1,1]]

Looking at many resources has confused me in the following:

  1. Will the Annoy first index these and then use to find nearest neighbors?
  2. Apply nearest neighbor tree and then index based on the neighbors? This seems to be the most appropriate one.. If it is then How does it index? I want to know the algorithm behind it..

Upvotes: 2

Views: 787

Answers (0)

Related Questions