A3C: The Policy Gradient Loss
Beginner
The problem, from first principles
The policy-gradient update raises the log-probability of actions that had positive advantage. Minimizing the negative of logp times advantage does exactly that when used with an optimizer.
From theory to code
Implement policy_gradient_loss(logp, adv), returning the negated mean of logp * adv.
Constraints
- Returns a Python float.
Hints
Hint 1
Multiply, average, negate.
Keywords
research-papersreinforcement-learningactor-criticasynchronous
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.