DDPG: Soft Target Updates
Beginner
The problem, from first principles
DDPG (Lillicrap et al., 2015) keeps slowly moving target networks for stability. Instead of copying the online weights every so often, the target takes a small step toward them after every update.
From theory to code
Implement soft_update(target, online, tau), returning the blended parameters.
Constraints
tauis small in practice, such as 0.001.
Hints
Hint 1
Blend the target and online arrays with weights 1 - tau and tau.
Keywords
research-papersreinforcement-learningcontinuous-controlactor-critic
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.