Skip to content

Latest commit

 

History

History
 
 

caql

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
README.md
README.md
 
 
agent_policy.py
agent_policy.py
 
 
agent_policy_test.py
agent_policy_test.py
 
 
caql_agent.py
caql_agent.py
 
 
caql_agent_test.py
caql_agent_test.py
 
 
caql_network.py
caql_network.py
 
 
caql_network_test.py
caql_network_test.py
 
 
dual_ibp_method.py
dual_ibp_method.py
 
 
dual_ibp_method_test.py
dual_ibp_method_test.py
 
 
dual_method.py
dual_method.py
 
 
dual_method_test.py
dual_method_test.py
 
 
epsilon_greedy_policy.py
epsilon_greedy_policy.py
 
 
epsilon_greedy_policy_test.py
epsilon_greedy_policy_test.py
 
 
gaussian_noise_policy.py
gaussian_noise_policy.py
 
 
gaussian_noise_policy_test.py
gaussian_noise_policy_test.py
 
 
policy.py
policy.py
 
 
replay_memory.py
replay_memory.py
 
 
replay_memory_test.py
replay_memory_test.py
 
 
run.sh
run.sh
 
 
train_eval.py
train_eval.py
 
 
utils.py
utils.py
 
 
utils_test.py
utils_test.py
 
 

README.md

CAQL, Continuous Action Q-Learning, is a class of algorithms for continuous-action Q-learning that can use several plug-and-play optimizers for the max-Q problem.

NOTE: MIP optimizer is not included in this initial version because the MIP optimizer is a Google-internal library and has not been open-sourced yet. However, it has a plan for open-source, and we will update this repository as soon as it is released.

For technical details of CAQL, refer to ICLR 2020 paper (https://openreview.net/forum?id=BkxXe0Etwr)