Ryan Julian comments

Results 37 comments of


                                            Ryan Julian

Pytorch Categorical GRU Policy

@krzentner I'm curious: what is your implementation plan for RNNs in torch/VPG? Is it to make the optimization trajectory-oriented (as in tf/VPG) or something else?

In pearl, qf1 and qf2 should be differently initialized

@JeongHyunho thank you for this issue report! This indeed looks like a bug, but I'd like to verify. Does `self._qf2` have the same parameters as `self._qf1`?

Rework logic for filling and checking replay buffer in torch sac, dog, and td3

`_train_once` is either private, or should be private, so I'm not sure what purpose this exception would have.

Docs page "Usage Guide -> Automatic hyperparameter tuning"

thanks @richardliaw !

Add "must have" features to make non-RL applications practical to write

These all look great to me and I look forward to reviewing the PRs! If you don't think they will all land around the same time, feel free to split...

update documentation on how to use rnns with tf/torch[pending]

If CG optimizer can't be used with RNNs (I don't think that's actually the case), we should detect that and raise an error.

Rework garage.torch.optimizers

Can you add a little bit more explanation for the design here? I'm concerned about using an ADT as the blanket input to policies, which makes the interface pretty complicated...

metaworld example for MT1 pick-place does not work

@avnishn

metaworld example for MT1 pick-place does not work

@guoyijie you can send us a tensorboard link by using the https://tensorboard.dev service. Do you mind trying MT1-reach? That would be a more reliable indictation of a possible problem.

tf/NPO is not compatible with max_episode_length=None

What version of TensorFlow is installed? `python -c 'import tensorflow as tf; print(tf.__version__)` within the Python environment you're running garage in will tell you. I'm guessing the problem may be...