tensorflow_similarity/models/contrastive_model.py (6 lines):
	- line 332: # TODO: figure out interesting metrics -- z Mae?
	- line 333: # TODO: check metrics are of the right type in compile?
	- line 371: # TODO: figure out interesting metrics -- z Mae?
	- line 372: # TODO: check metrics are of the right type in compile?
	- line 600: # TODO (ovallis): Refactor the following indexing code into a MixIn.
	- line 821: TODO: more detailed explaination.


tensorflow_similarity/indexer.py (2 lines):
	- line 107: # FIXME support custom objects
	- line 503: FIXME: more detailed explanation.


tensorflow_similarity/models/similarity_model.py (2 lines):
	- line 242: # TODO (ovallis): Refactor the following indexing code into a MixIn.
	- line 456: TODO: more detailed explaination.


tensorflow_similarity/evaluators/memory_evaluator.py (1 line):
	- line 224: # TODO (ovallis): Assert if index is empty, or if the lookup is empty.


tensorflow_similarity/api/__init__.py (1 line):
	- line 35: 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.


tensorflow_similarity/losses/simclr.py (1 line):
	- line 14: # FIXME original reference


tensorflow_similarity/visualization/projector.py (1 line):
	- line 115: # FIXME: 2d vs 3d