Parameters

Customizable parameters

resign_threshold

• range: (0.0 <= x <= 1.0)
• default: 0.20
• Resign when winning rate is lower than this value.

expand_threshold

• range: (1 <= x)
• default: 15
• Expand search tree node how many times visited.

c_puct

• range: (0.0 < x)
• default: 2
• a constant determining the level of exploration
• 

value_scale

• range: (0.0 <= x <= 1.0)
• default: 0.8
• Mix the value network and rollout evaluations
• 

policy_temperature

• range: (0.0 <= x)
• default: 0.49
• Softmax temperature of policy network

policy_temperature_inc

• range: (0.0 <= x)
• default: 0.056

tgr1_rate

• range: (0 <= x <= 100)
• default: 0
• Percentage of playing cached moves is tree search in simulation.

lgrf1_rate

• range: (0 <= x <= 100)
• default: 0
• Percentage of playing Last-Good-Reply-Forgetting moves in simulation.