Dueling Networks: Centering the Advantages
Beginner
The problem, from first principles
Before combining with V, the advantages are centered so they have zero mean across actions. This is the piece that pins down the value and advantage streams.
From theory to code
Implement centered_advantage(A), subtracting the mean of the advantage vector.
Constraints
- The output sums to zero.
Hints
Hint 1
Subtract A.mean() from every entry.
Keywords
research-papersreinforcement-learningdeep-q-learningarchitectures
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.