How to implement different deep Q models

tensorflow / agents

TF-Agents: A reliable, scalable and easy to use TensorFlow library for Contextual Bandits and Reinforcement Learning.

Apache License 2.0

2.78k stars 718 forks source link

How to implement different deep Q models #257

Open wsun899 opened 4 years ago

wsun899 commented 4 years ago

The input of the current implementation of deep Q model in the tf-agents is observation and its output is the Q values of its possible actions. I need a different structure of deep Q model: Input: observation + action, and output: the Q value of the action. I am just wondering if there are any examples or suggestions for quick implementation on the framework. Thanks,

sguada commented 4 years ago

Take a look at https://github.com/tensorflow/agents/blob/master/tf_agents/agents/ddpg/critic_network.py

wsun899 commented 4 years ago

Thanks a lot, Sguada!