## Constants
RENDER_CHARTS = True
#################
# Setup agents #
#################
players[0] = A2CAgent(
action_space=132,
observation_space=env.observation_space,
n_latent_var=128,
K_epochs=4,
gamma=0.999,
network_save_name='/agents/A2CAgent/saved_models/A2C_test_2',
network_load_name=None)
names[0] = "A2C Agent"
players[1] = random_actions(env.num_actions_per_turn, 1, map_name)
names[1] = 'Random Agent'
#################
actions = {}
## Set high episode to test convergence
# Change back to resonable setting for other testing
n_episodes = 2500
#########################
# Statistic variables #
#########################
k = 50 # Used for average win rates
p = 5 # Print episodic results every p episodes
stats = AgentStatistics(names[0],
map_name=map_name,
train=False,
network_save_name=None,
network_load_name=None,
)
names[0] = "DQN Agent"
# Create an array of all agents that could be used during training
opposing_agents = [
{
'name': 'Random Agent Delay',
'agent': random_actions_delay(env.num_actions_per_turn, 1, map_name)
},
{
'name': 'Random Agent',
'agent': random_actions(env.num_actions_per_turn, 1, map_name)
},
{
'name': 'Bull Rush',
'agent': bull_rush(env.num_actions_per_turn, 1)
},
{
'name': 'All Cycle',
'agent': all_cycle(env.num_actions_per_turn, 1)
},
{
'name': 'Base Rush v1',
'agent': base_rushV1(env.num_actions_per_turn, 1)
},
{
'name': 'Cycle Rush Turn 25',