def main(w_count, success_count):
env = MyEnv({})
while True:
# print(f'step {step}')
# ランダムアクションの選択
# action_index = env.action_space.sample()
a = np.array([0, 1, 2, 3])
p = np.array([1, 1, 1, 1])
"""
a = np.array([0, 1, 2, 3, 4, 5, 6, 7, 8])
p = np.array([1, 1, 1, 1, 1, 1, 1, 1, 1])
"""
p = p / np.sum(p)
action_index = np.random.choice(a, p=p)
# 環境を１step 実行
observation, reward, done, _ = env.step(action_index)
if args.render_mode == 'human':
print(f'\naction is selected at {env.steps}')
status_print(env, observation, reward, done)
# 環境の描画
shot = env.render(mode=args.render_mode)
# Space keyでpause, デバッグ用
pause_for_debug()
# エピソードの終了処理
if done:
# print('done')
w_count, success_count = conunt_results(env, w_count,
success_count)
break
return w_count, success_count
def main():
# Initialize ray
ray.init(ignore_reinit_error=True, log_to_driver=False)
# Generate & Check environment
env = MyEnv({})
# Define trainer agent
model_name = MODEL_NAME
config = ppo.DEFAULT_CONFIG.copy()
config['env_config'] = {}
config['num_gpus'] = 0
config['framework'] = 'tfe'
config['eager_tracing'] = True
agent = ppo.PPOTrainer(config=config, env=MyEnv)
agent.restore(model_name)
for idx in range(90):
""" Initialization """
observation = env.reset()
frames = []
""" Save some initial values """
fighter_0 = env.fighter.ingress
jammer_0 = env.jammer.ingress
while True:
action_index = agent.compute_action(observation)
# 環境を１step 実行
observation, reward, done, _ = env.step(action_index)
# 環境の描画とビデオ録画
# shot = env.render(mode=args.render_mode)
frames.append(env.render(mode=args.render_mode))
# Space keyでpause, デバッグ用
pause_for_debug()
# Slow down rendering
pygame.time.wait(10)
# エピソードの終了処理
if done:
status_print(env, observation, reward, done, fighter_0,
jammer_0)
video_name = ALGORITHM + '_' + env.mission_condition + '-' + str(
idx)
make_video(video_name, frames)
make_jason(env, video_name, fighter_0, jammer_0)
break
def main():
# Initialize ray
ray.init(ignore_reinit_error=True, log_to_driver=False)
# Generate & Check environment
env = MyEnv({})
# Define trainer agent
model_name = MODEL_NAME
config = ppo.DEFAULT_CONFIG.copy()
config['env_config'] = {}
config['num_workers'] = NUM_WORKERS
config['num_gpus'] = 0
config['framework'] = 'tfe'
config['eager_tracing'] = True
agent = ppo.PPOTrainer(config=config, env=MyEnv)
agent.restore(model_name)
success_history = []
success_count = 0
for idx in range(N_EVAL_EPISODES):
""" Initialization """
observation = env.reset()
frames = []
""" Save some initial values """
fighter_0 = env.fighter.ingress
jammer_0 = env.jammer.ingress
while True:
action_index = agent.compute_action(observation)
# 環境を１step 実行
observation, reward, done, info = env.step(action_index)
# 環境の描画とビデオ録画
# shot = env.render(mode=args.render_mode)
frames.append(env.render(mode=args.render_mode))
# Slow down rendering
# pygame.time.wait(10)
# エピソードの終了処理
if done:
success_history.append(info['success'])
if info['success'] > .5:
success_count += 1
break
n_success = success_count
n_fail = N_EVAL_EPISODES - n_success
if np.sum(success_history) != success_count:
raise Exception('Something is wrong!')
""" Summarize results """
print('==================== Summary of the results ====================')
print(
f'Mission conditions = w1 : w2 : w3 = '
f'{env.mission_probability[0]:.3f} : {env.mission_probability[1]:.3f} : {env.mission_probability[2]:.3f}'
)
print(f' Model is < {MODEL_NAME} >')
print(
f' Number of success missions: {round(n_success)} / {N_EVAL_EPISODES}, '
f' Number of failed missions {round(n_fail)} / {N_EVAL_EPISODES}')