Global similarity scores

The similarity.cosine module provides tools for calculating global similarity scores between query and database. This is done using cosine similarity between fixed-length embeddings extracted by neural networks.

similarity.cosine

CosineSimilarity

Wraps cosine similarity to be usable in SimilarityPipeline.

__call__(query, database, **kwargs)

Calculates cosine similarity given query and database feature datasets.

Parameters:

Name	Type	Description	Default
query	FeatureDataset	Query dataset of deep features.	required
database	FeatureDataset	Database dataset of deep features.	required

Returns:

Name	Type	Description
similarity	array	2D numpy array with cosine similarity.

cosine_similarity(a, b)

Calculate cosine similarity between two sets of vectors. Pytorch Equivalent to sklearn.metrics.pairwise.cosine_similarity.

Example

query and database are instance of FeatureDataset with deep features.

from wildlife_tools.similarity import CosineSimilarity

similarity = CosineSimilarity()
sim = similarity(query, database)