def TrainSmallAgent(NrGames, writeOn, savePath, show=False):
from SmallAgent import SmallAgent as Agent
#make the enviroment
env = Enviroment(fullImage=False, enhancedReward=True, autoFire=True)
if (show):
env.reset()
env.render()
agent = Agent()
#see how many times to save the agent
NrBatches = int(NrGames / float(writeOn)) + 1
#Train some games then save the results keep repeating until all number of games have been played
for i in range(NrBatches):
print "batch", i, "of", NrBatches
#train the agent
reward, steps = trainAgent(env, agent, writeOn, show=show)
#save the itermediate agent
agent.save(savePath + "_" + str(i * writeOn))
#save the stats
f = open(savePath, 'a')
for r, s in zip(reward, steps):
f.write(str(r) + "," + str(s) + '\n')
f.close()
env.makeVideo(savePath + "_" + str(i * writeOn) + ".gif")
def TrainLargeAgent(NrGames, writeOn, savePath, show=False):
from LargeAgent import LargeAgent as Agent
#make the enviroment
env = Enviroment(enhancedReward=True, autoFire=True)
#because of the influence of tensorflow draw the enviroment once before making any call to tensorflow of don't draw at all
if (show):
env.reset()
env.render()
#make the agent
agent = Agent()
#see how many times to save the agent
NrBatches = int(NrGames / float(writeOn)) + 1
totalNrFrames = 0
#Train some games then save the results keep repeating until all number of games have been played
for i in range(NrBatches):
print "batch", i, "of", NrBatches
#train the agent
reward, steps = trainAgent(env, agent, writeOn, show=show)
totalNrFrames += np.sum(steps)
#save the itermediate agent
agent.save(savePath + "_" + str(i * writeOn))
#save the stats
f = open(savePath, 'a')
for r, s in zip(reward, steps):
f.write(str(r) + "," + str(s) + '\n')
f.close()
env.makeVideo(savePath + "_" + str(i * writeOn) + ".gif")
'w'))
json.dump(
self.actionTable,
open("./" + folder + '/' + filename + '/' + filename + "act.table",
'w'))
# "./" + folder + '/' + filename + '/' + filename)
if (__name__ == "__main__"):
import time
from Eviroment import Enviroment
env = Enviroment(gameMode="vector2", rewardMode="follow", autoFire=True)
agent = DoraAgent(double=True)
action = 0
frame = -1
skipFrame = 1
done = False
S = env.reset()
while not done:
frame += 1
if (frame % skipFrame == 0):
action = agent.getAction(S)
S, r, done, info = env.step(action)
time.sleep(0.5)
env.render()
print "game evaluation: ", env.reward
from Eviroment import Enviroment
import numpy as np
if(__name__ == "__main__"):
from SmallAgent import SmallAgent as Agent
show = False
#make the enviroment
env = Enviroment(fullImage=False,enhancedReward=True,autoFire=True)
if(show):
env.reset()
env.render()
agent = Agent.load('')
write = 0
rweards = []
for i in range(500):
S = env.reset()
done = False
while not done:
action = agent.getAction(S)
Snew,reward,done,info = env.step(action)
agent.update(S,action,Snew,reward,done)
S = Snew
if(show):
env.render()
if(write == 0):
write = 1
env.makeVideo('./randomMovement.gif')