Superlight weight vector searcher, implemented using KD Trees. data is saved to json and the tree stucture is saved to the .sav structure. Examples of how to add and load data are in the notebooks. It's good for tiny applications. Uses embeddings from Sentence embedding package.
add_one()
takes the form of
query: Query to pass in as example
embedding_query: Query with mask [MASK] to create embedding out of
data: The sequel
Generates the three additional columns:
embedding: Creates embedding from BERT from the embedding_query
index: (automatically +1 last element)
hash: hash of the embedding_query
KDTree_ask.sav is for ask table indexs/KDTree
KDTree_predict.sav and KDTree_visualise.sav ^ infer