GAE: Discounted Returns
Intermediate
The problem, from first principles
The return from a step is the total discounted reward collected from there on. It is the quantity policy-gradient methods weight their updates by, in the simplest (REINFORCE) form.
From theory to code
Implement discounted_returns(rewards, gamma), computing the discounted return from each step.
Constraints
- Compute from the last step backward.
Hints
Hint 1
Start at zero after the final step and add r_t + gamma times the return from the next step.
Keywords
research-papersreinforcement-learningpolicy-gradientadvantage
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.