COSIE.utils.construct_knn_graph_hnsw

construct_knn_graph_hnsw(data, k=20, space='l2')

Efficiently compute approximate k-nearest neighbor (k-NN) graph using the hnswlib library. This method is suitable for large-scale datasets.

Parameters

datatorch.Tensor

A tensor of shape (n_cells, n_features) representing the input feature matrix.

kint, optional

Number of nearest neighbors to retrieve for each sample. Default is 20.

spacestr, optional

Distance metric used to build the index. Must be one of {‘l2’, ‘ip’, ‘cosine’}.

• ‘l2’: Euclidean distance

• ‘ip’: Inner product

• ‘cosine’: Cosine similarity

Default is ‘l2’.

Returns

edge_indextorch.LongTensor

A tensor of shape (2, n_edges) representing the edges of the graph. Each column represents an edge from source node to target node.