search

TheButlah / makrl

makrl - modular algorithm kit for reinforcement learning
4 stars 1 forks source link

Implement a Dueling Q Network as a `Model` #16

Closed TheButlah closed 5 years ago

TheButlah commented 5 years ago

The dueling q network looks like a promising improvement on basic Q learning. We should implement this as a subclass of ActionModel and then allow the QLearning agent to use it as its model based on an argument to the constructor.

TheButlah commented 5 years ago

Assign yourself when you start to work on this

TheButlah commented 5 years ago

This is not on the table for now. Our priorities rn are to complete some baseline models instead of complex ones.