google / trax / 4365

21%

master: 21%

Build Type

push

travis-ci

Committed by Copybara-Service

Commit Message

An assortment of QWR improvements.

- Multiple value network synchronizations in one epoch.
- Softmax and logsumexp Q-value aggregation.
- LayerNorm observation normalization as in the ACME framework.
- Learnable standard deviation in the Gaussian distribution - either separate or shared across dimensions.

PiperOrigin-RevId: 333323682

Run Details

52 of 52 new or added lines in 4 files covered. (100.0%)

2433 of 11353 relevant lines covered (21.43%)

0.21 hits per line