def chaser_greedy(env_name, episode_counts, display=True, fps=10):
env = gym.make(env_name)
chaser = GreedyAgent(actions_n=4)
epi = 0
for epi in range(episode_counts):
state = env.reset()
steps = 0
while True:
if display:
time.sleep(1 / fps)
env.render()
action = chaser.act(state)
state_, _, done, _ = env.step(action)
steps += 1
state = state_
if done:
print('Episode: %d\tSteps: %d' % (epi + 1, steps))
break
# Initiate the renderer
env_renderer = RenderTool(
local_env,
gl="PILSVG",
agent_render_variant=AgentRenderVariant.ONE_STEP_BEHIND,
show_debug=False,
screen_height=1200, # Adjust these parameters to fit your resolution
screen_width=1800) # Adjust these parameters to fit your resolution
######### Get arguments of the script #########
parser = argparse.ArgumentParser()
parser.add_argument("-step", type=int, help="steps")
args = parser.parse_args()
######### Custom controller: imports #########
controller = GreedyAgent(218, local_env.action_space[0])
######### Define custom controller #########
def my_controller(obs, number_of_agents, astar_paths_readable, timestamp):
_action = {}
grid = obs[0][0]
# Chose an action for each agent in the environment
for a_id in range(number_of_agents):
status, position, direction, initial_position, target, speed = obs[
a_id][1]
action = None
if status == RailAgentStatus.ACTIVE:
next_cell = None
if astar_paths_readable[a_id] is not None:
screen_width=1800) # Adjust these parameters to fit your resolution
######### Get arguments of the script #########
parser = argparse.ArgumentParser()
parser.add_argument("-a", type=int, help="id of the agent to be displayed")
parser.add_argument("-step", type=int, help="steps")
args = parser.parse_args()
######### Select agent #########
show_agents = [0]
if args.a:
show_agents[0] = int(args.a)
print(show_agents)
######### Custom controller setup #########
controller = GreedyAgent(218, env.action_space[0])
# Build graph from transition map
print("\nCompute transition graph from generated rail grid.")
graph_generator = GraphBuilder(env.rail.width, env.rail.height, env.rail.grid)
# create whitelist, which cells not to remove from graph
cell_whitelist = set()
for a_id in show_agents:
ag = env.agents[a_id]
start = graph_generator.convert_indexes_2_node(ag.initial_position, )
if ag.position is not None:
start = graph_generator.convert_indexes_2_node(ag.position)
cell_whitelist.update(
[start, graph_generator.convert_indexes_2_node(ag.target)])
config = {'players': 2, 'verbose': False}
EPISODES = 100
DISCARDS = [.1 * i for i in range(11)]
PLAYS = [.5 + .05 * i for i in range(11)]
max_rewards = []
mean_rewards = []
median_rewards = []
for play_threshold in PLAYS:
max_reward = []
mean_reward = []
median_reward = []
for base_discard in DISCARDS:
agents = [GreedyAgent(config) for i in range(2)]
for agent in agents:
agent.play_threshold = play_threshold
agent.discard_thresholds = [
base_discard + i * (play_threshold - base_discard) / 9
for i in range(9)
]
runner = GameRunner(agents, EPISODES, config)
runner.run()
rewards = runner.rewards
max_reward.append(np.max(rewards))
mean_reward.append(np.mean(rewards))
median_reward.append(np.median(rewards))
max_rewards.append(max_reward)
mean_rewards.append(mean_reward)
median_rewards.append(median_reward)
screen_height=1200, # Adjust these parameters to fit your resolution
screen_width=1800) # Adjust these parameters to fit your resolution
######### Get arguments of the script #########
parser = argparse.ArgumentParser()
parser.add_argument("-agent", type=int, help="id of the agent to be displayed")
parser.add_argument("-step", type=int, help="steps")
args = parser.parse_args()
######### Select agent #########
show_agents = [2]
if args.agent is not None:
show_agents[0] = int(args.agent)
######### Custom controller setup #########
controller = GreedyAgent(218, env.action_space[0])
my_grid = [[Node((i, j), env.rail.grid[i, j]) for j in range(env.rail.width)]
for i in range(env.rail.height)]
astar_planner = AStarAgent(my_grid, env.rail.width, env.rail.height)
astar_paths_readable = [None for _ in range(env.number_of_agents)]
# run A* for the selected agent
for a_id in show_agents:
ag = env.agents[a_id]
start = ag.initial_position
if ag.position is not None:
start = ag.position
start = Node(start, env.rail.grid[start[0], start[1]], dir=ag.direction)
end = Node(ag.target, env.rail.grid[ag.target[0], ag.target[1]])
astar_paths_readable[a_id] = [
node.point for node in astar_planner.aStar(start, end)