Summary: 86 instances, 73 unique

Text	Count
# TODO: calculate log_prob for l2 normalization	1
# TODO: can we get this working well with jupyter?	1
# TODO: do priority sampling with torch as well.	1
# FIXME: Doesn't support LRScheduler yet	1
# TODO: we currently use the same data for test and validation.	2
# TODO (@czxttkl): use when we introduce padding	1
TODO: remove this file once we can infer everything.	2
# FIXME: this is a hack around https://github.com/PyTorchLightning/pytorch-lightning/pull/9360	1
# TODO for future cleanup: kind of a misnomer now, since not really "difference"	1
# TODO: write this for OSS	7
# TODO: handle possible actions/mask here	1
# TODO: have an embedder here	1
has_user_feat: bool = False  # TODO: deprecate	1
minibatches_per_step (optional, TODO: currently unused): the number of minibatch updates	1
# TODO: handle value/mask of DocList	1
// TODO: Implement pid controller to replace this fixed shift	1
TODO: maybe replace with sparse-to-dense tensor function?	1
TODO: made environment easier to learn from by not using RecSim.	1
# TODO add Discrete Single Step Synthetic Reward Predictor	1
# TODO: consider possible_actions_mask	2
# TODO: Probably should create a new model type	1
TODO: implement for stack size > 1	2
# TODO: make this take in a	1
# TODO: calls to _maybe_run_optimizer removed, should be replaced with Trainer parameter	1
# TODO: Create a generic framework for type conversion	1
# TODO: rename underlying function to get_max_possible_values_and_idxs	1
# TODO: export these as observable values	2
# TODO: Implement ExtraData.from_dict	1
# TODO: log prob is affected by clamping, how to handle that?	1
# TODO: check to ensure no rl parameter value is set that isn't actively used by class	1
"sequence_number": 0,  # TODO: Support sequences	1
# TODO: remove this dependency	1
# TODO: make this a property of EnvWrapper?	1
# TODO: should this be in base class?	1
# TODO: Add below get_data_module() method once methods in	1
# TODO: why is reward net commented out?	1
# TODO: remove RecsimObsPreprocessor and move it here	1
# TODO: finish	1
# TODO: abs value to make probability?	1
TODO: replace with redis (python) and hiredis (C) for better RASP support	1
# TODO: this wouldn't work for possible_actions_mask (list of value, presence)	1
# TODO: Move this to appropriate location	1
// TODO: Implement CDF of t-distribution for bayesian UCB	1
# TODO: T62502977 create learnable feature vectors for start symbol	1
# TODO assert regarding offsets length compared to value	1
# TODO: raise this bar after training stabilize	1
# TODO: Load tensors from torch files.	1
# TODO: rename "observation" to "state" in Transition and return cls(**d)	1
# TODO: Warmstarts means and vars using previous solutions (T48841404)	1
TODO: change this to a deterministic subsample.	1
# FIXME: calculate model_metrics_values when q_network_cpe is added	1
TODO: make this an embedder.	1
# TODO: make abstract	2
# TODO: can this line be hit currently in ReAgent?	1
# FIXME: hardcoded for now	1
# TODO: We probably should put member vars into local vars to	1
# TODO: Just use kwargs here?	2
# FIXME: this only works for one-hot encoded actions	1
FIXME: Remove this function after the issue above is resolved	1
# TODO: make return serving feature data	1
# TODO: make generic get_action_idxs for each trainer class	1
# FIXME: These config types are misplaced but we need to write FBL config adapter	1
# TODO: add more methods to simplify gym code	1
)  # TODO: if score cap not needed, deprecate	1
# FIXME: model_values and model_metrics_values should be	1
TODO: This function should return ReAgentLightningModule &	1
# TODO: Must setup (or mock) trainer and a LoggerConnector to call self.log()!	1
# TODO: Add option for simple modulo and other hash functions	1
# FIXME: epoch argument is not really correct	1
# TODO: write SlateQ Wrapper	1
# TODO: Save tensors to torch files.	1
# TODO: for OSS	1
# TODO: add in critic	1