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")
json.dump(
self.dimensionTable,
open("./" + folder + '/' + filename + '/' + filename + "dim.table",
'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)
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')
return debugAgent()
"""
nothing to save so do nothing
"""
def save(self,filename):
print self.gameStates
np.savetxt(filename +"/X.csv",np.asarray(self.gameStates),delimiter=",")
np.savetxt(filename +"/Y.csv",self.labels,delimiter=",")
if(__name__ == "__main__"):
from Eviroment import Enviroment
import time
env = Enviroment(Breakout=True,gameMode="vector",rewardMode="label",autoFire=True)
agent = debugAgent()
batch=0
done = False
for game in range(100):
S = env.reset()
done = False
while not done:
action = agent.getAction(S)
S1,r,done,info = env.step(action)
agent.update(S,action,S1,r,done, batch)
S = S1
batch+=1
if (__name__ == "__main__"):
#The number game to train the agent on
NrOfTrainingGames = 50000
#save the result after every number game
writeOn = 10000
mode = sys.argv[1]
saveFile = sys.argv[2]
if (mode == "small"):
double = sys.argv[3] == "true"
trajectory = sys.argv[4] == "true"
games = [
Enviroment(gameMode="vector", rewardMode="label", autoFire=True),
Enviroment(Breakout=False,
gameMode="vector",
rewardMode="label",
autoFire=True),
Enviroment(gameMode="vector", rewardMode="label", autoFire=True)
]
gameNames = ["Breakout_1", "Pong", "Breakout_2"]
experimentOnSmall(saveFile, games, gameNames, double, trajectory)
elif (mode == "large"):
games = [
Enviroment(gameMode="neural", rewardMode="label", autoFire=True),
Enviroment(Breakout=False,
gameMode="neural",
rewardMode="label",
autoFire=True),
# run Pong sim.
import experiment
from Eviroment import Enviroment
from DoraModelTime import DoraAgent as Agent
import numpy as np
import matplotlib as mat
mat.use("AGG")
from matplotlib import pyplot as plt
import pdb
# create the agent and load the saved data.
agent = Agent(double=False)
#actionTable = json.load(open("./" + folder + '/' + filename + '/' + filename + "act.table",'r'))
#dimensionTable = json.load(open("./" + folder + '/' + filename + '/' + filename + "dim.table",'r'))
#agent.actionTable = actionTable
#agent.dimensionTable = dimensionTable
agent = agent.load('Breakout_1_800')
agent.flipWeights("X", "Y")
#
enviroment = Enviroment(Breakout=False,
gameMode="vector",
rewardMode="follow",
autoFire=True)
NrGames = 200
writeOn = 10
savePath = 'Pong1'
experiment.testAgent(enviroment, agent, NrGames, writeOn, savePath, show=True)