Hybrid Online and Offline Reinforcement Learning for Tibetan Jiu Chess

Table 1

The experimental parameters.

Toolkit package	CNTK 2.7
Runtime	.Net 4.7.2
Operating system	Windows 10
Central processing unit	AMD 2700X@4.0 GHz
Random access memory	32 GB
Graphics processing unit	RTX2070
Threads used	16
Development environment	Visual Studio 2017 Community
	0.1
Step	200
Learning rate	0.001
(Q-learning)	0.628
(SARSA)	0.372