Summary: 14 instances, 10 unique

Text	Count
The default implementation can scale up to medium deployment (1M-10M+ points) easily, provided the computers have enough memory. For very large scale deployments you will need to sublcass the compoments to match your own architetctue. See FIXME colab to see how to deploy TensorFlow Similarity in production.	1
# TODO: figure out interesting metrics -- z Mae?	2
FIXME: more detailed explanation.	1
# TODO: check metrics are of the right type in compile?	2
# FIXME: 2d vs 3d	1
TODO: more detailed explaination.	2
# FIXME support custom objects	1
# FIXME original reference	1
# TODO (ovallis): Assert if index is empty, or if the lookup is empty.	1
# TODO (ovallis): Refactor the following indexing code into a MixIn.	2