google-deepmind / google-deepmind/reverb

Batching writer.append steps

Open
#78 7 comments 0 reactions 0 assignees View on GitHub
Dominant language
C++
Stars
793
Forks
113
PR merge metrics
No merged PRs in 30d

Description

Hey there. Is it possible to batch the `writer.append` command (e.g. extend) in Reverb? I am currently doing something like the following:

```
for exp in exps:
self._writer.append(exp)
```
Where I loop over 400 steps with quite large observation sizes. I am also using Launchpad. It seems that this append is taking about 20% of the episode computation time. Is there something like writer.extend in Reverb or some other fancy method to maybe do only one write for a batch of steps for improved performance? Should I maybe try and just skip the `writer.append` and directly try and create trajectories from the concatenated episode steps? Thanks.

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.