import numpy as np
import matplotlib.pyplot as plt
if __name__ == "__main__":
max_steps = 8000
num_runs = 1
# Create and pass agent and environment objects to RLGlue
environment = WindygridEnvironment()
agent = SarsaAgent()
rlglue = RLGlue(environment, agent)
del agent, environment # don't use these anymore
for run in range(num_runs):
episode=[]
time_step=[]
rlglue.rl_init()
while True:
rlglue.rl_episode()
time_step.append(rlglue.num_steps())
episode.append(rlglue.num_episodes())
if rlglue.num_steps() > 8000:
break
plt.plot(time_step,episode,label="8 actions")
plt.xticks([0, 1000, 2000, 3000, 4000, 5000, 6000, 7000, 8000])
plt.xlabel('Time steps')
plt.ylabel('Episode', rotation=90)
plt.legend(loc=2)
plt.show()
# save average value function numpy object, to be used by plotting script
from rl_glue import RLGlue
from windy_env import WindyEnvironment
from sarsa_agent import SarsaAgent
import numpy as np
import matplotlib.pyplot as plt
max_steps = 8000
steps = 0
episodes = 0
ep_list = []
step_list = []
environment = WindyEnvironment()
agent = SarsaAgent()
rl = RLGlue(environment, agent)
rl.rl_init()
while steps < max_steps:
rl.rl_episode(max_steps)
steps = rl.num_steps()
episodes = rl.num_episodes()
# print(steps, episodes)
ep_list.append(episodes)
step_list.append(steps)
plt.xlabel('Time steps')
plt.ylabel('Episodes')
plt.plot(step_list, ep_list)
plt.show()
from rl_glue import RLGlue
from windy_env import WindyEnvironment
from n_step_sarsa_agent import SarsaAgent
import numpy as np
import time
import matplotlib.pyplot as plt
if __name__ == "__main__":
start_time = time.time()
max_steps = 8000
# Create and pass agent and environment objects to RLGlue
environment = WindyEnvironment()
agent = SarsaAgent()
rlglue = RLGlue(environment, agent)
del agent, environment # don't use these anymore
rlglue.rl_init()
L1 = []
L2 = []
n = rlglue.rl_agent_message('n')
a = rlglue.rl_agent_message('a')
while rlglue.num_steps() < max_steps:
L1.append(rlglue.num_steps())
rlglue.rl_episode(10000)
episodes = rlglue.num_episodes()
L2.append(episodes)
plt.title(str(n) + '-step sarsa with ' + str(a) + " actions")
plt.plot(L1, L2)
plt.show()
rlglue = RLGlue(environment, agent)
rlglue.rl_agent_message('n = 0')
del agent, environment # don't use these anymore
steps1 = {}
L1 = []
L2 = [0]*num_episodes
for episode in range(num_episodes):
steps1[episode]=[]
L1.append(episode+1)
for run in range(num_runs):
np.random.seed(run)
rlglue.rl_init()
step = 0
for episode in range(num_episodes):
rlglue.rl_episode(max_steps)
new_step = rlglue.num_steps()
steps1[episode].append(new_step-step)
step = new_step
for episode in range(num_episodes):
L2[episode] = sum(steps1[episode])/num_runs
plt.plot(L1[1:],L2[1:],label = '0 planning steps')
environment =GridEnvironment()
agent = DynaqAgent()
rlglue = RLGlue(environment, agent)
rlglue.rl_agent_message('n = 5')
del agent, environment # don't use these anymore
steps1 = {}
L1 = []
L2 = [0]*num_episodes
for episode in range(num_episodes):
from sarsa_agent import SarsaAgent
import numpy as np
if __name__ == "__main__":
max_total_step = 8000
total_num_episode = [0]
current_step = [0]
# Create and pass agent and environment objects to RLGlue
environment = WindEnvironment()
agent = SarsaAgent()
rlglue = RLGlue(environment, agent)
del agent, environment # don't use these anymore
# set seed for reproducibility
np.random.seed(1)
# initialize RL-Glue
rlglue.rl_init()
# loop for the experiment step less that 8000
while (rlglue.num_steps() < max_total_step):
rlglue.rl_episode(max_total_step)
total_num_episode.append(rlglue.num_episodes())
current_step.append(rlglue.num_steps())
print(total_num_episode)
np.savez('windy.npz', timeSteps=current_step, Episodes=total_num_episode)
# to load:
# data = np.load('windy.npz')
# print(data["timeSteps"]) to get the array