acme

acme copied to clipboard

When I run the cql algorithm, I found the algorithm only execute behavior clone. I checked the config used. The training step is 100 and the 'num_bc_iters' is set to...

jiangjiadi

Adds a more informative samples per second logging for `reverb_dataset.py`. Before: ``` Iteration 0 finished in 0.8698790073394775s Iteration 1 finished in 0.9794516563415527s Iteration 2 finished in 0.9291434288024902s ``` After: ```...

hartikainen

Reverb Replay Buffers + Vectorized/Batched Environments

1

I'm trying to use a reverb replay buffer with a batched environment like 'envpool' where the api returns a batch of experience whenever the either `.reset` or `.step ` is...

wbrenton

Is the last layer in distributional critic optimized (jax implementation)?

1

Hi! I was wondering if the last layer in the distributional critic used in dmpo, for example, is optimized in the jax implementation, since it is instantiated in the __call__...

Jogima-cyber

Reference numbers for `reverb_benchmark.py`

2

Hey, I'm setting up a custom ACME learning algorithm and currently the learning seems to be quite heavily bottlenecked by the reverb dataset sampling. I tried to understand the reverb...

hartikainen

InferenceServer Error

13

I am trying to use seel-RL style centralized server for inference. I modified the IMPALA example by adding the `inference_server.InferenceServerConfig`, which is passed to `experiments.make_distributed_experiment`. I keep getting the below...

kinalmehta

Library broken

1

The package does not run any more. I tried installing it in multiple ways on my computer (pip, git clone,etc) and nothing seems to work. How should we install this...

adiaconu11

examples seem to be broken - examples/baselines/rl_continuous

2

Installed dm-acme using the command `pip install dm-acme[envs,jax,tf] ` Running examples fails with a message that it can't import that package. I go into the python environment to view the...

brian1985

rewritten jnp.squeeze(self._value_layer(inputs), axis=-1)

1

jnp.squeeze(self._value_layer(inputs), axis=-1) can be rewritten as self._value_layer(inputs)[..., 0] to make the code look more concise.

skywalker0803r

[`video.py`](https://github.com/deepmind/acme/blob/d1e69c92000079b118b868ce9303ee6d39c4a0b6/acme/wrappers/video.py#L29) imports `matplotlib` which is not an explicit dependency listed in `setup.py`. Happy to contribute a fix but I do not know which solution do you guys want: 1. Explicitly...

cemlyn007