google-research / google-research/dice_rl

lagrangian estimation of the policy value (average reward) in the neural_dice.py

Open
#8 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
112
Forks
16
PR merge metrics
No merged PRs in 30d

Description

The lagrangian estimation of the policy value (average reward) in the neural_dice.py is computed as

lagrangian = nu_zero + self._norm_regularizer * self._lam + constraint

But according to the paper, I think it should be

lagrangian = nu_zero + dual_step+ self._norm_regularizer * self._lam + constraint

which includes the dual estimate.

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.