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")
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")
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')
if(i == 250 and write == 1):
write = 2
env.makeVideo('./halfRandomMovement.gif')
rweards.append(env.reward)
print "End Game",env.reward
env.makeVideo('./minimalRandomMovement.gif')
first = np.mean(rweards)
second = np.std(rweards)
rweards = []
for _ in range(30):
S = env.reset()
done = False
def load(filename):
return randomAgent()
"""
nothing to save so do nothing
"""
def save(self, filename):
pass
if (__name__ == "__main__"):
from Eviroment import Enviroment
env = Enviroment(Breakout=True,
enhancedReward=True,
fullImage=False,
autoFire=True)
agent = designedAgent()
done = False
S = env.reset()
while not done:
action = agent.getAction(S)
S, r, done, info = env.step(action)
env.render()
env.makeVideo('designedAgent.gif')
