def __init__(self, gamma=0.9, alpha=0.1, epsilon=0.1, num_episodes=500):
self.q_table = pd.DataFrame(0,index=pd.MultiIndex.from_product\
([[0,1,2,3],[0,1,2,3]]),columns=['UP','DOWN','LEFT','RIGHT'])
self.alpha = alpha
self.gamma = gamma
self.epsilon = epsilon
self.num_episodes = num_episodes
self.env1 = env.ENV()
self.action_count, self.actions = self.env1.getActionItems()
def __init__(self,
gamma=0.9,
alpha=0.1,
epsilon=0.1,
num_episodes=500,
storage_value=[],
row_num=5,
col_num=5):
self.col_list = list(range(0, col_num))
self.row_list = list(range(0, row_num))
self.q_table = pd.DataFrame(0,index=pd.MultiIndex.from_product\
([self.row_list,self.col_list]),columns=['UP','DOWN','LEFT','RIGHT'])
self.alpha = alpha
self.gamma = gamma
self.epsilon = epsilon
self.num_episodes = num_episodes
self.env1 = env.ENV(storage_value, row_num, col_num)
self.action_count, self.actions = self.env1.getActionItems()
def test(self):
try:
q_table = pd.read_pickle(
'q_table.pkl') #read q table from the database saved
except:
txt.delete("1.0", "end")
txt.insert(END, 'Try to train the Agent before testing it\n')
return
if (q_table is not None):
env1 = env.ENV()
#display_environment(env1)#display the init state of the environment
done = False
while not done: #until it reaches the destination
state = env1.getAgentPosition() #get current position
action = q_table.loc[state, :].index[q_table.loc[
state, :].values.argmax(
)] #fetch the next best possible action from the Q table learnt
_, _, done = env1.step(action) #execute the action
def handle_click(event):
env1 = env.ENV()
update(env1)
time.sleep(1)
env1.step('RIGHT')
update(env1)
time.sleep(1)
env1.step('RIGHT')
update(env1)
time.sleep(1)
env1.step('LEFT')
update(env1)
time.sleep(1)
env1.step('DOWN')
update(env1)
time.sleep(1)
env1.step('DOWN')
update(env1)
time.sleep(1)
env1.step('DOWN')
update(env1)
time.sleep(1)
env1.step('DOWN')
update(env1)
time.sleep(1)
env1.step('RIGHT')
update(env1)
time.sleep(1)
env1.step('RIGHT')
update(env1)
import time
import numpy as np
import config
import data_loader
import env
import agent
import policy
DL = data_loader.DataLoader()
DL.read_data(split='test')
env = env.ENV(DL)
greedy = policy.MaskedGreedyPolicy()
multi = policy.MaskedMultiDisPolicy()
agent = agent.pgAgent(env=env,
nb_warm_up=400,
policy=multi,
testPolicy=greedy,
gamma=0.95,
lr=0.0001,
memory_limit=2000,
batchsize=64,
train_interval=200)
time1 = time.time()
for round in range(1000):
print("------------------------------------------------------")
print('\n\n train ' + str(round) + '_' + str((time.time() - time1) / 60))
def reset_environment(self):
#just creating another instance of environment variable
del (self.env1)
self.env1 = env.ENV()
return self.env1.getAgentPosition()
def reset_environment(self, storage_value, row_num, col_num):
#just creating another instance of environment variable
del (self.env1)
self.env1 = env.ENV(storage_value, row_num, col_num)
return self.env1.getAgentPosition()