from Normal import exploration
from Normal import measure_length
from Normal import moving_average
#load data
double_without=pickle.load(open( "double_without_record.p", "rb"))
double_record=pickle.load(open( "double_record.p", "rb"))
double_evaluation_record_without=pickle.load(open( "double_evaluation_without_record.p", "rb"))
double_evaluation_record=pickle.load(open( "double_evaluation_record.p", "rb"))
evalu_without_ave=sum(double_evaluation_record_without)/evaluation_episode_number
evalu_ave=sum(double_evaluation_record)/evaluation_episode_number
#plot training results
def moving_average(x, w):
return np.convolve(x, np.ones(w), 'valid') / w
double_without_average=moving_average(double_without,average_over)
double_record_average=moving_average(double_record,average_over)
fig=plt.figure(figsize=(13,7))
env_standard=800
x=range(len(double_record_average))
plt.plot(x,double_without_average,label='Normal Training\nEvaluation %s'%evalu_without_ave,color='black',linestyle='-.')
plt.plot(x,double_record_average,label='Coached by PID Controller\nEvaluation %s'%evalu_ave,color='magenta')
plt.xlabel('Episode Number', fontsize='large')
plt.ylabel('Episode Reward', fontsize='large')
plt.legend(loc='upper left',ncol=1, borderaxespad=0,prop={'size': 18})
plt.axhline(y=env_standard, color='black', linestyle='dotted')
plt.savefig('double.png')
actions=[thigh_actions,leg_actions,foot_actions,left_thigh_actions,left_leg_actions,left_foot_actions]
states, terminal, reward = environment.execute(actions=actions)
reward= -1
episode_reward+=reward
agent.observe(terminal=terminal, reward=reward)
else:
states, terminal, reward = environment.execute(actions=actions)
agent.observe(terminal=terminal, reward=reward)
episode_reward+=reward
record.append(episode_reward)
reward_record[k][i]=record
temp=np.array(record)
reward_record_average[k][i]=moving_average(temp,average_over)
#evaluate
episode_reward = 0.0
eva_reward_record=[]
print('evaluating agent with boundary position at %s and prohibitive parameter %s' %(prohibition_position[k],prohibition_parameter[i]))
for j in tqdm(range(evaluation_episode_number)):
episode_reward=0
states = environment.reset()
internals = agent.initial_internals()
terminal = False
while not terminal:
actions, internals = agent.act(states=states, internals=internals, independent=True, deterministic=True)
states, terminal, reward = environment.execute(actions=actions)
episode_reward += reward
eva_reward_record.append(episode_reward)
ip_without = pickle.load(open("ip_without_record.p", "rb"))
ip_record = pickle.load(open("ip_record.p", "rb"))
ip_evaluation_record_without = pickle.load(
open("ip_evaluation_without_record.p", "rb"))
ip_evaluation_record = pickle.load(open("ip_evaluation_record.p", "rb"))
evalu_without_ave = sum(
ip_evaluation_record_without) / evaluation_episode_number
evalu_ave = sum(ip_evaluation_record) / evaluation_episode_number
def moving_average(x, w):
return np.convolve(x, np.ones(w), 'valid') / w
ip_without_average = moving_average(ip_without, average_over)
ip_record_average = moving_average(ip_record, average_over)
fig = plt.figure(figsize=(13, 7))
env_standard = 800
x = range(len(ip_record_average))
plt.plot(x,
ip_without_average,
label='Normal Training\nEvaluation %s' % evalu_without_ave,
color='black',
linestyle='-.')
plt.plot(x,
ip_record_average,
label='Coached by PID Controller\nEvaluation %s' % evalu_ave,
color='magenta')
plt.xlabel('Episode Number', fontsize='large')