Yicheng Luo

Annotation with collections.abc.Mapping does not work with python3.8.

Fix policy loss gradient in TD3

1

The gradients dq_da is currently incorrect. The gradients for each dimension from the action should be summed instead of averaged as per https://github.com/deepmind/rlax/blob/master/rlax/_src/policy_gradients_test.py#L55 For example, the D4PG agent also doesn't...

BUGFIX: Unpin tensorflow_datasets

3

The current pinned version does not work for some of the d4rl locomotion v2 datasets due to the wrong specification in metadata types.

Add py.typed

1

This makes type annotations available to users. Otherwise, all types from Acme are currently typed as Any

Improving snapshotting in JAX distributed experiments

8

Hi, I have started adopting `acme.jax.experiments.make_distributed_experiment` for training distributed agents with LaunchPad. Right now, it seems that there is no way for me to configure snapshotting, which is a useful...

Is the priority computation correct in prioritized DQN

4

Hi, Cross posting a related issue here https://github.com/deepmind/reverb/issues/91 After looking at the DQN agent implementation in Acme, I am not sure if the prioritized variant matches the original PER paper....

Hi Acme developers, I would like to point out some issues with the current examples. Right now, the instructions on the README install the CPU version of JAX by default....

Providing nightly builds

1

Hi, Thanks for the awesome library! It would be great if acme can provide nightly builds (similar to tf-nightly and tfp-nightly) which would allow the users to pin the versions...

Specify non-pytree node dataclass fields

3

Hi, Thanks for making this awesome library! Is it possible to specify fields in the chex.dataclass definitions to not include certain fields? This is a feature supported in flax https://flax.readthedocs.io/en/latest/_modules/flax/struct.html#dataclass...

Add cholesky and gradients

6

This change is [](https://reviewable.io/reviews/tensorflow/tfjs-core/1492)