Dueling Networks: The Greedy Action Is Unchanged
Beginner
The problem, from first principles
Because V is the same for every action in a state, adding it cannot change which action has the largest Q. The greedy choice is therefore decided by the advantages alone.
From theory to code
Implement dueling_argmax(V, A), returning the greedy action of the dueling Q-function.
Constraints
- Ties choose the lowest index.
Hints
Hint 1
Build the dueling Q-values, then take argmax.
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.