gnes.indexer.chunk.annoy module

class gnes.indexer.chunk.annoy.AnnoyIndexer(num_dim, data_path, metric='angular', n_trees=10, *args, **kwargs)[source]

Bases: gnes.indexer.base.BaseChunkIndexer

Initialize an AnnoyIndexer

  • num_dim (int) – when set to -1, then num_dim is auto decided on first .add()
  • data_path (str) – index data file managed by the annoy indexer
  • metric (str) –
  • n_trees (int) –
  • args
  • kwargs
add(keys, vectors, weights, *args, **kwargs)[source]

adding new chunks and their vector representations

  • keys (List[Tuple[int, Any]]) – list of (doc_id, offset) tuple
  • vectors (ndarray) – vector representations
  • weights (List[float]) – weight of the chunks

Declare class attributes/members that can not be serialized in standard way

query(keys, top_k, *args, **kwargs)[source]
Return type:List[List[Tuple]]
train(*args, **kwargs)

Train the model, need to be overrided