Summary: 41 instances, 36 unique

Text	Count
# TODO: Update to also compute m2c_align_mat and m2t_align_mat	1
# TODO (PVP) still a bit hacky here - there might be a better solution	2
# TODO: sum the probability of all occurrences	1
# TODO fix	1
if column.type in ["number", "time"]: # TODO fine-grained date	1
# TODO: Fix this	1
# TODO (PVP): this should later be handled by the forward fn() in each model in the future see PR 3140	2
# TODO batching	2
if "name" in kwargs: del kwargs["name"] #TODO: fix this	1
# TODO: Write 'train', 'val', 'test' somewhere else	1
# TODO initialize using xavier	1
# TODO: words should match memory	3
# TODO: Mask other probabilities first?	1
# TODO: from_cond should be true from non-bert model	1
# TODO batching	1
# TODO: Record that this constructor is a sequence fragment?	1
assert factorize_sketch == 2 #TODO support other grammars	1
# TODO: Support multiple layers	1
#TODO: should get types from the data	1
# TODO: In Torch 1.3, discontinue use of torch.jit.Attribute so that	1
# TODO: Figure out how to get right sizes (query, key) to module	1
# TODO: the dropout mask should be stored in the state instead?	1
# TODO: not nice	1
# TODO: better heuristic e.g., tables that have exact match	1
# TODO: Support batch_first	1
## TODO: Don't hardcode train	1
TODO:	1
# TODO: support slices	1
# TODO: Get rid of surrounding quotes	1
# TODO: smarter mapping.	1
# TODO initialize using xavier	1
# TODO: Figure out how to get right sizes (query, value) to module	1
# TODO: This should be automatically inferred from encoder	1
#TODO: from_cond should be true from non-bert model	1
# TODO: make it posible to deal with "singleton?"	1
# TODO: Check that things don't blow up when some lengths are 0	1