def main():
MAP = map_8
map_size = MAP[1]
beta = 0.999
env = FrozenLakeEnv(desc=MAP[0], is_slippery=True)
print("Game Map:")
env.render()
start_time = time.time()
v, pi = sync_policy_iteration(env, beta=beta)
v_np, pi_np = np.array(v), np.array(pi)
end_time = time.time()
run_time['Sync Policy Iteration'] = end_time - start_time
print("time:", run_time['Sync Policy Iteration'])
print_results(v, pi, map_size, env, beta, 'sync_pi_gs')
from copy import deepcopy
temp_dict = deepcopy(run_time)
print("All:")
for _ in range(len(temp_dict)):
min_v = float('inf')
for k, v in temp_dict.items():
if v is None:
continue
if v < min_v:
min_v = v
name = k
temp_dict[name] = float('inf')
print(name + ": " + str(min_v))
print()
def main():
MAP = map_DH
map_size = MAP[1]
simulator = FrozenLakeEnv(desc=MAP[0])
epsilon = 0.3 # Select 0.3 or 0.05
learning_rate = 0.001 # Select 0.001 or 0.1
learning_episodes = 30000 # 30000 (DH) or 100000 (16)
test_interval = 100
do_test = True
sarsa_agent = SARSAagent(simulator,
epsilon,
learning_rate,
learning_episodes,
map_size,
test_interval=test_interval,
do_test=do_test)
start_time = time.time()
total_rewards, q_table = sarsa_agent.learn_and_evaluate()
run_time['SARSA agent'] = time.time() - start_time
print("Learning time:\n")
print(run_time['SARSA agent'])
if do_test:
plot_result(total_rewards, sarsa_agent.test_interval,
[sarsa_agent.agent_name])
plot_image(q_table, MAP[0], map_size)
def main():
MAP = map_8
map_size = MAP[1]
beta = 0.999
env = FrozenLakeEnv(desc=MAP[0], is_slippery=True)
print("Game Map:")
env.render()
start_time = time.time()
v, pi = sync_value_iteration(env, beta=beta)
v_np, pi_np = np.array(v), np.array(pi)
end_time = time.time()
run_time['Sync Value Iteration'] = end_time - start_time
print("time:", run_time['Sync Value Iteration'])
print_results(v, pi, map_size, env, beta, 'sync_vi_gs')
ray.shutdown()
ray.init(include_webui=False,
ignore_reinit_error=True,
redis_max_memory=500000000,
object_store_memory=5000000000)
beta = 0.999
env = FrozenLakeEnv(desc=MAP[0], is_slippery=True)
print("Game Map:")
env.render()
start_time = time.time()
v, pi = distribured_value_iteraion(env, beta=beta, workers_num=4)
v_np, pi_np = np.array(v), np.array(pi)
end_time = time.time()
run_time['Sync distributed VI'] = end_time - start_time
print("time:", run_time['Sync distributed VI'])
print_results(v, pi, map_size, env, beta, 'dist_vi')
from copy import deepcopy
temp_dict = deepcopy(run_time)
print("All:")
for _ in range(len(temp_dict)):
min_v = float('inf')
for k, v in temp_dict.items():
if v is None:
continue
if v < min_v:
min_v = v
name = k
temp_dict[name] = float('inf')
print(name + ": " + str(min_v))
print()
def learn_and_evaluate(self, MAP):
workers_id = []
for _ in range(self.cw_num):
simulator = FrozenLakeEnv(desc=MAP[0])
collecting_worker.remote(self.server, simulator, self.epsilon)
for _ in range(self.ew_num):
worker_id = evaluation_worker.remote(self.server, simulator)
workers_id.append(worker_id)
ray.wait(workers_id, len(workers_id))
return ray.get(self.server.get_reuslts.remote())
def main():
ray.shutdown()
ray.init(include_webui=False,
ignore_reinit_error=True,
redis_max_memory=500000000,
object_store_memory=5000000000,
temp_dir='~/ray_tmp')
MAP = map_DH
map_size = MAP[1]
simulator = FrozenLakeEnv(desc=MAP[0])
simulator.reset()
epsilon = 0.3 # Fix epsilon as 0.3
learning_rate = 0.001 # Select 0.001 or 0.1
learning_episodes = 30000 # 30000 (DH) or 100000 (16)
test_interval = 100
batch_size = 100
do_test = True
start_time = time.time()
distributed_ql_agent = distributed_QL_agent(simulator,
epsilon,
learning_rate,
learning_episodes,
map_size,
test_interval=test_interval,
batch_size=batch_size,
cw_num=8,
ew_num=4,
do_test=do_test)
total_rewards, q_table = distributed_ql_agent.learn_and_evaluate(MAP)
run_time['Distributed Q-learning agent'] = time.time() - start_time
print("Learning time:\n")
print(run_time['Distributed Q-learning agent'])
if do_test:
plot_result(total_rewards, test_interval,
[distributed_ql_agent.agent_name])
plot_image(q_table, MAP[0], map_size)
w_id = VI_worker.remote(VI_server, data_id, finish_worker,
start_state, end_state)
workers_list.append(w_id)
start_state += batch_size
if start_state > S:
start_state = S
end_state += batch_size
if end_state > S:
end_state = S
error = ray.get(VI_server.get_error_and_update.remote())
v, pi = ray.get(VI_server.get_value_and_policy.remote())
return v, pi
beta = 0.999
env = FrozenLakeEnv(desc=MAP[0], is_slippery=True)
print("Game Map:")
env.render()
start_time = time.time()
v, pi = sync_value_iteration_distributed(env, beta=beta, workers_num=4)
v_np, pi_np = np.array(v), np.array(pi)
end_time = time.time()
run_time['Sync dis'] = end_time - start_time
print("time:", run_time['Sync distributed v2'])
print_results(v, pi, map_size, env, beta, 'dist_vi')