Summary: 97 instances, 94 unique

Text	Count
# TODO: upon getting rid of pytorch 1.4, kill this	1
TODO: make all subclasses compatible with DataBrowser	1
# TODO: ImageLoader functionality. Is from comment_battle,	1
# TODO: get rid of longest_label	1
nn.ReLU(),  # TODO: should this also be gelu?	1
# TODO: add in zero3. https://github.com/facebookresearch/ParlAI/issues/3753	1
# TODO: blueprint type strings need to be imported here to register the blueprints -	1
# TODO: 'bad_workers' is for compatibility. Before, it was only non-empty if a	1
supported_architects: ClassVar[List[str]] = ["mock"]  # TODO update	1
# TODO: will refactor this at some point soon most likely	1
# TODO: try to deprecate this, preferring straight txt2vec	1
# TODO List:	1
# TODO: merge this function in with the make_world in in parlai/crowdsourcing/tasks/model_chat/worlds.py	1
# TODO: mark as abstract?	3
# TODO: make it so that we don't necessarily have to evaluate	1
# TODO: proper train/test/valid splits	1
# TODO: speed these calculations up	1
# TODO: not so sure about this...	1
# TODO: We possibly can log these errors,	1
TODO: implement this	1
# TODO: support self._validate_self_observe_invariants() method of TorchAgent	1
# TODO: mark as optionally abstract?	1
# TODO: enable it back. This leads lr decay to 0.	1
# TODO: Awkward to modify the input opt but needed to use	1
# TODO: merge for the initialization step	1
# TODO: eventually all teachers should return Messages, so	1
# TODO: test the internal zoo.	1
# TODO: PCA	1
'--eval-dynamic-batching',  # FIXME: see https://github.com/facebookresearch/ParlAI/issues/3367	1
# TODO: set some of these args with opt	1
# FIXME: see issue tracked in https://github.com/facebookresearch/ParlAI/issues/3367	1
# TODO: Shall we use smaller valid/test sets? Even 10% is A LOT here	1
# TODO: copy	1
time.sleep(THREAD_SLEEP)  # TODO: use threading.Event() rather than polling	1
# TODO: remove on the final version	1
# TODO: Look into possibly overriding opt('dict_file') with new path	1
// TODO handle v. rare cases of message drops - should send timeout status	1
# TODO: move this over to TorchRankerAgent when it is ready.	1
# TODO DEPRECATIONDAY: FbDialogueTeacher is being deprecated, should we	1
// TODO: add support for radio input type	1
# TODO: switch to ~/.parlai/	1
# TODO: Sorting the batch will result in various local metrics being broadcasted	2
# TODO: PCA + RP	1
# TODO: all teachers should eventually create messages	1
# TODO: possible need a fixed seed here in the future	1
# TODO: remove assert in the final version	1
# TODO modularize the shared components to dedup the code	1
# TODO: rename the folders nicer so they make more sense, maybe even have	1
TODO: the final chat data is currently stored both in	1
NOTE: Preference for new projects is Dict output (see the TODO below).	1
# TODO: support more weight tying options?	1
'category': "Somewhere",  # TODO pull from DB?	1
# TODO: in `self._check_output_key()`, there is other logic for ignoring	1
# TODO support data-based attachments?	1
# TODO: Get rid of this hack	1
# TODO: remove typing of self after switching to pytest regressions, in which we	1
# TODO load this data in using DataBrowser	1
# TODO: once the analysis scripts are fully switched over to DataBrowser, remove	1
# TODO: add token dropout?	1
# TODO: initialize biases to 0	1
# TODO: support self._validate_observe_invariants() method of TorchAgent	1
# TODO: if we get rid of the streaming datafile num_episodes/num_examples	1
# TODO: add reverse conversation as well	1
# TODO: won't need this after we have proper logging levels set	1
TODO: Only support Dict. Deprecate ` pd.DataFrame` when no other code is relying on it.	1
TODO: Determine how to do this	1
// TODO move constants to props rather than state	1
// TODO: replace this hacky solution for removing remaining message from onboarding	1
# TODO: add dropout	1
# TODO: remove and replace with multiteachers only?	1
# TODO: check to ensure the agent has self_observe	1
# scale_parameter=True, TODO: enable it back. This leads lr decay to 0.	1
# TODO: replace direct folder access with a call to	1
# TODO: mark as abstract, get rid of runtime error.	1
# TODO Can we grab these from somewhere in the built content?	1
# TODO: I don't think we need to re-chunk this actually; just iterate	1
TODO: this can probably be moved to a method of an abstract test class once all	1
# TODO: deprecate override_opts	1
observation = Message(observation)  # TODO: eventually this will not be	1
# TODO: remove type ignore with pytorch 1.5:	1
# TODO: update the server (and then this) for batch responses.	1
# TODO: remove this attribute once chat data is only stored in the Mephisto	1
# TODO: should check that utterances specified are all bot	1
TODO: perhaps have the image chat task make use of this context information	1
to load a large image feature dict in memory). #TODO Could be the default option	1
# TODO: maybe generalize this	1
// TODO move to CSS	1
# TODO: this could almost certainly be faster with a regex \s+ -> ' '	1
# TODO: mark as abstract	1
# TODO: more thoroughly test out whether one of these choices is best and add a	1
# TODO: move this to Torch Ranker Agent	1
# Resume TODO: isfile() may take too long ?? Should I try in a .tmp file	1
# TODO: must avoid having these in the dataset in the first place	1
# TODO: Migration plan: TorchAgent currently supports being passed	1