pytorch/examples

Actor critic example not using discount rate properly

Open

#744 opened on Mar 27, 2020

View on GitHub
 (3 comments) (0 reactions) (0 assignees)Python (21,634 stars) (9,429 forks)batch import
good first issuetriaged

Description

The Actor Critic example (which is actually an implementation of REINFORCE-with-baseline as pointed out in https://github.com/pytorch/examples/issues/573), does not use the discount rate properly.

The loss should include \gamma ^ t, as shown in the box on page 330 of Sutton & Barto:

image

Contributor guide

Tech stack
pythonpytorch
Domain
machine learningai
Issue type
bug
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-3 hours
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
clear
Prerequisites
REINFORCE algorithmPyTorch basicsDiscount factor concept
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
45
Research direction
Examine the actor critic example file (likely in the reinforcement learning folder). The current code likely computes the loss without weighting by gamma^t. Refer to the referenced issue #573 and the Sutton & Barto equation. The fix involves multiplying each log probability term by gamma^t in the policy gradient loss calculation. Check the comments in the issue for any additional guidance.