def launch_environment(parameters):
"""Start the sumo-rlglue environment
(This function is executed in a separate process using
multiprocessing.)
"""
import rl_glue_sumo_environment
environment = rl_glue_sumo_environment.SumoEnvironment()
EnvironmentLoader.loadEnvironment(environment)
def main():
EnvironmentLoader.loadEnvironment(
InvasiveEnvironment(
simulationParameterObj = None,
actionParameterObj = None,
Bad_Action_Penalty = -10000,
fixedStartState = False,
nbrReaches = REACHES,
habitatSize = HABITATS,
seed = 1))
description=
'Run a POMDP problem file as a domain in RL-Glue in network mode.')
group = parser.add_mutually_exclusive_group(required=True)
group.add_argument(
"--file",
type=str,
default=config_files[0],
help="Run POMDP domain given the path to a POMDP problem file.")
group.add_argument("--list",
action='store_true',
default=False,
help="List path to included POMDP problem files.")
args = parser.parse_args()
if args.list:
print "Included POMDP problem files:"
for file in config_files:
print file
else:
EnvironmentLoader.loadEnvironment(
POMDPEnvironment(spec_filename=args.file))
parser = argparse.ArgumentParser(
description='Run a specified POMDP in RL-Glue in network mode.')
parser.add_argument("--pomdp_file",
type=str,
help="Filename for POMDP spec file to load and use.",
required=True)
args = parser.parse_args()
EnvironmentLoader.loadEnvironment(POMDPEnvironment(args.pomdp_file))
returnRO = Reward_observation_terminal()
returnRO.r = theReward
returnRO.o = theObs
returnRO.terminal = episodeOver
return returnRO
def env_cleanup(self):
pass
def env_message(self,inMessage):
return "I don't know how to respond to your message";
def addGridworldArgs(parser):
parser.add_argument("--size_x", type=float, default=10, help="Size of the gridworld in the x (horizontal) dimension, where 1.0 is the unit of movement.")
parser.add_argument("--size_y", type=float, default=10, help="Size of the gridworld in the y (vertical) dimension, where 1.0 is the unit of movement.")
parser.add_argument("--goal_x", type=float, default=10, help="Goal x coordinate")
parser.add_argument("--goal_y", type=float, default=10, help="Goal y coordinate")
parser.add_argument("--noise", type=float, default=0, help="Standard deviation of additive noise to generate")
parser.add_argument("--fudge", type=float, default=1.4143, help="Distance from goal allowed before episode is counted as finished")
parser.add_argument("--random_restarts", type=bool, default=False, help="Randomly assign x,y initial locations.")
if __name__=="__main__":
import argparse
parser = argparse.ArgumentParser(description='Run 2D Noisy Continuous Gridworld environment in network mode.')
addGridworldArgs(parser)
args = parser.parse_args()
EnvironmentLoader.loadEnvironment(Gridworld(size_x=args.size_x, size_y=args.size_y, goal_x=args.goal_x, goal_y=args.goal_y, noise=args.noise, random_start=args.random_restarts, fudge=args.fudge))
self.o.doubleArray=list(range(0,50000))
terminal=0
if self.stepCount==200:
terminal=1
ro=Reward_observation_terminal()
ro.r=1.0
ro.o=self.o
ro.terminal=terminal
return ro
self.o.intArray=list(range(0,5))
#cheating, might break something
self.o.doubleArray=list(range(0,5))
terminal=0
if self.stepCount==5000:
terminal=1
ro=Reward_observation_terminal()
ro.r=1.0
ro.o=self.o
ro.terminal=terminal
return ro
def env_cleanup(self):
pass
def env_message(self,inMessage):
return None;
if __name__=="__main__":
EnvironmentLoader.loadEnvironment(test_speed_environment())
def printState(self): numRows=len(self.map) numCols=len(self.map[0]) print "Agent is at: "+str(self.agentRow)+","+str(self.agentCol) print "Columns:0-10 10-17" print "Col ", for col in range(0,numCols): print col%10, for row in range(0,numRows): print print "Row: "+str(row)+" ", for col in range(0,numCols): if self.agentRow==row and self.agentCol==col: print "A", else: if self.map[row][col] == self.WORLD_GOAL: print "G", if self.map[row][col] == self.WORLD_MINE: print "M", if self.map[row][col] == self.WORLD_OBSTACLE: print "*", if self.map[row][col] == self.WORLD_FREE: print " ", print if __name__=="__main__": EnvironmentLoader.loadEnvironment(mines_environment())
result = re.match('set (.+) (.+)', msg)
if msg.startswith('set'):
param, value = msg.split(None, 2)[1:]
self.debug('set', param, value)
self.env_message_set_param(param, value)
elif msg.startswith('get'):
param = msg.split(None, 1)[1]
return self.env_message_get_param(param)
else:
return self.env_message_handler(msg)
def debug(self, *args):
""" Print a debug msg """
if self.debug:
args = [str(a) for a in args]
print "%s: %s" % (self.name, ' '.join(args))
def step_out(self, *args):
if self.output_steps:
args = [str(a) for a in args]
print ' '.join(args)
if __name__ == '__main__':
#p = PuddleEnvironment()
#p.env_start()
EnvironmentLoader.loadEnvironment(PuddleEnvironment())
return (-1 - i)
if self.presentCol > 6999 and self.presentCol < 8000:
if self.presentRow > 5999 and self.presentRow < 9000:
return -1
if self.presentCol > 5999 and self.presentCol < 7000:
if self.presentRow > 4999 and self.presentRow < 8000:
return -2
#Reward
if self.presentCol > 10999:
if self.presentRow < 1000:
return 10
return 0
#Checking if the current position is the goal state
def goalcheck(self):
# Only A goal state
if self.presentCol > 10999:
if self.presentRow < 1000:
return True
return False
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(puddle_world())
r = 0.0
if np.any(self.state_ranges[:,0] > self.state[:]) or \
np.any(self.state_ranges[:,1] < self.state[:]):
# r = -1
r = -np.sum(3.0 * self.state_ranges[:, 1]**2)
r *= 6000 - self.num_sim_steps
terminate = True
else:
# perr = np.linalg.norm(self.prevState[:2] - self.state_goal[:2])
# nerr = np.linalg.norm(self.state[:2] - self.state_goal[:2])
# r = math.exp(-np.sum(abs(self.state[:2]-self.state_goal[:2])/(.1*(self.state_ranges[:2,1]-self.state_ranges[:2,0]))))* \
# math.exp(-np.sum(abs(self.state[3:5]-self.state_goal[3:5])/(.1*(self.state_ranges[3:5,1]-self.state_ranges[3:5,0]))))* \
# math.exp(-np.sum(abs(self.state[6:8]-self.state_goal[6:8])/(.1*(self.state_ranges[6:8,1]-self.state_ranges[6:8,0]))))
# r = math.exp(-np.sum(abs(self.state[:2]-self.state_goal[:2])/(.1*(self.state_ranges[:2,1]-self.state_ranges[:2,0]))))* \
# math.exp(-np.sum(abs(self.state[3:5]-self.state_goal[3:5])/(.1*(self.state_ranges[3:5,1]-self.state_ranges[3:5,0]))))
# r -= (np.sum(((self.state[:2]-self.state_goal[:2])/(self.state_ranges[:2,1]-self.state_ranges[:2,0]))**2)+ \
# np.sum(((self.state[3:5]-self.state_goal[3:5])/(self.state_ranges[3:5,1]-self.state_ranges[3:5,0]))**2))
r -= (self.state[0] - self.state_goal[0])**2
r -= (self.state[1] - self.state_goal[1])**2
r -= self.state[3]**2
r -= self.state[4]**2
terminate = False
print("reward " + str(r))
return r, terminate
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(vrep_environment())
theReward=-1 episodeOver=1 if self.currentState >= 20: self.currentState=20 theReward=1 episodeOver=1 theObs=Observation() theObs.intArray=[self.currentState] returnRO=Reward_observation_terminal() returnRO.r=theReward returnRO.o=theObs returnRO.terminal=episodeOver return returnRO def env_cleanup(self): pass def env_message(self,inMessage): if inMessage=="what is your name?": return "my name is dqn_environment, Python edition!"; else: return "I don't know how to respond to your message"; if __name__=="__main__": EnvironmentLoader.loadEnvironment(dqn_environment())
default=0,
help="Standard deviation of additive noise to generate")
parser.add_argument(
"--fudge",
type=float,
default=1.4143,
help="Distance from goal allowed before episode is counted as finished"
)
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser(
description=
'Run 2D Noisy Continuous Gridworld environment in network mode.')
addTaxiArgs(parser)
args = parser.parse_args()
fuelloc = None if args.fuel_loc[0] < 0 else args.fuel_loc
walls = numpy.array(args.wall) if args.wall is not None else None
landmarks = numpy.array(
args.landmark) if args.landmark is not None else None
EnvironmentLoader.loadEnvironment(
Taxi(args.size_x,
args.size_y,
walls=walls,
landmarks=landmarks,
fuel_loc=fuelloc,
fickleness=args.fickleness,
noise=args.noise,
fudge=args.fudge))
returnRO.r = theReward
returnRO.o = theObs
returnRO.terminal = episodeOver
return returnRO
def env_cleanup(self):
pass
def env_message(self,inMessage):
return "I don't know how to respond to your message";
@register_environment
class MountainCar(MountainCarND):
name = "Mountain Car"
def __init__(self, **kwargs):
kwargs['dimension'] = 2
super(MountainCar, self).__init__(**kwargs)
if __name__=="__main__":
import argparse
parser = argparse.ArgumentParser(description='Run Noisy Mountain Car environment in network mode.')
parser.add_argument("--noise", type=float, default=0, help="Standard deviation of additive noise to generate, affects the action effects.")
parser.add_argument("--random_restarts", type=bool, default=False, help="Restart the cart with a random location and velocity.")
args = parser.parse_args()
EnvironmentLoader.loadEnvironment(MountainCar(noise=args.noise, random_start=args.random_restarts))
"""
MENACE agent environment wrapper
"""
import sys
# Ugly!
sys.path.append('../')
from agents.symmetry_agent import SymmetryAgent
from wrapper_environment import WrapperEnvironment
from rlglue.environment import EnvironmentLoader
class SymmetryEnvironment(WrapperEnvironment, SymmetryAgent):
name = 'symmetry_agent'
player = 2
def env_play(self):
action = self.do_step(self.state)
self.state = action.intArray
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(SymmetryEnvironment())
self.map[i:i + self.n_cols])) + '\n'
self.history.append(current_map)
# 試合の様子を記録
if rot.r == self.game.r_lose:
f = open('history.txt', 'a')
history = '\n'.join(self.history)
f.writelines('# START\n' + history + '# END\n\n')
f.close()
# 決着がついた場合は agentのagent_end
# 決着がついていない場合は agentのagent_step に続く
return rot
def env_cleanup(self):
pass
def env_message(self, message):
pass
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Deep Q-Learning')
parser.add_argument('--size',
'-s',
default=6,
type=int,
help='Reversi board size')
args = parser.parse_args()
EnvironmentLoader.loadEnvironment(KmoriReversiEnvironment(args.size))
episodeOver=0
theReward=0
theObs=Observation()
theObs.intArray=np.zeros(50816)
returnRO=Reward_observation_terminal()
returnRO.r=theReward
returnRO.o=theObs
returnRO.terminal=episodeOver
return returnRO
def env_cleanup(self):
pass
def env_message(self,inMessage):
if inMessage=="what is your name?":
return "my name is dqn_environment, Python edition!";
else:
return "I don't know how to respond to your message";
if __name__ == '__main__':
App = TetrisApp()
#App.run()
EnvironmentLoader.loadEnvironment(TetrisApp())
print('main start')
type=float,
default=5.,
help="Payment received per unit product sold.")
parser.add_argument("--cost",
type=float,
default=2.,
help="Cost per unit product purchased.")
parser.add_argument("--discount_factor",
type=float,
default=0.999,
help="Discount factor to learn over.")
parser.add_argument(
"--time_period",
type=int,
default=20,
help="Time period for problem. (Number of steps to run)")
parser.add_argument(
"--noise",
type=float,
default=0,
help="Standard deviation of additive noise to generate")
args = parser.parse_args()
EnvironmentLoader.loadEnvironment(
BatchReplenishment(demand_mean=args.demand_mean,
demand_std=args.demand_std,
payoff=args.payoff,
cost=args.cost,
gamma=args.discount_factor,
time_period=args.time_period,
noise=args.noise))
def checkCurrentTerminal(self):
return self.checkTerminal(self.agentRow, self.agentCol)
def updatePosition(self, theAction):
# When the move would result in hitting an obstacles, the agent simply doesn't move
newRow = self.agentRow
newCol = self.agentCol
newRow += self.FIXED_DISTANCE * math.cos(theAction)
newCol += self.FIXED_DISTANCE * math.sin(theAction)
# Check if new position is out of bounds or inside an obstacle
if self.checkValid(newRow, newCol):
self.agentRow = newRow
self.agentCol = newCol
return False
else:
return True
def calculateReward(self, hitBoundary):
if hitBoundary:
return -0.5
if distance.euclidean([self.agentRow, self.agentCol], self.END_STATE) < 0.5 * self.FIXED_DISTANCE:
return 10.0
return 0.0
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(threeroom_environment())
if self.reward_noise > 0:
theReward += numpy.random.normal(scale=self.reward_noise)
theObs = Observation()
theObs.doubleArray = self.state.tolist()
returnRO = Reward_observation_terminal()
returnRO.r = theReward
returnRO.o = theObs
returnRO.terminal = episodeOver
return returnRO
def env_cleanup(self):
pass
def env_message(self,inMessage):
return "I don't know how to respond to your message";
if __name__=="__main__":
import argparse
parser = argparse.ArgumentParser(description='Run Noisy Acrobot environment in network mode.')
parser.add_argument("--noise", type=float, default=0, help="Standard deviation of additive noise to generate, affects the action effects.")
parser.add_argument("--random_restarts", type=bool, default=False, help="Restart the state with random values.")
args = parser.parse_args()
EnvironmentLoader.loadEnvironment(Acrobot(noise=args.noise, random_start=args.random_restarts))
return returnRO
def env_cleanup(self):
pass
def env_message(self,inMessage):
return "I don't know how to respond to your message";
def addTaxiArgs(parser):
parser.add_argument("--size_x", type=float, default=5, help="Size of the gridworld in the x (horizontal) dimension, where 1.0 is the unit of movement.")
parser.add_argument("--size_y", type=float, default=5, help="Size of the gridworld in the y (vertical) dimension, where 1.0 is the unit of movement.")
parser.add_argument("--landmark", action="append", nargs=2, help="Add a landmark, give x y coordinates", type=float)
parser.add_argument("--wall", type=float, action="append", nargs=2, help="Add a wall, give x coordinate and size in y with sign indicating starting at the bottom (+) or top (-)")
parser.add_argument("--fuel_loc", type=float, default=[2.0, 1.0], nargs=2, help="x y coordinate of the fuel station")
parser.add_argument("--fickleness", type=float, default=0, help="Probability of the passenger changing their destination mid-route.")
parser.add_argument("--noise", type=float, default=0, help="Standard deviation of additive noise to generate")
parser.add_argument("--fudge", type=float, default=1.4143, help="Distance from goal allowed before episode is counted as finished")
if __name__=="__main__":
import argparse
parser = argparse.ArgumentParser(description='Run 2D Noisy Continuous Gridworld environment in network mode.')
addTaxiArgs(parser)
args = parser.parse_args()
fuelloc = None if args.fuel_loc[0] < 0 else args.fuel_loc
walls = numpy.array(args.wall) if args.wall is not None else None
landmarks = numpy.array(args.landmark) if args.landmark is not None else None
EnvironmentLoader.loadEnvironment(Taxi(args.size_x, args.size_y, walls=walls, landmarks=landmarks, fuel_loc=fuelloc, fickleness=args.fickleness, noise=args.noise, fudge=args.fudge))
action='store',
type='string',
default='./rlglue_param.json',
help="json file of simulation parameters")
parser.add_option('-d',
'--dst',
dest='savepath',
action='store',
type='string',
default='./result/data',
help="Save data path.")
(options, args) = parser.parse_args()
#print 'options', options
if not options.test:
EnvironmentLoader.loadEnvironment(ElecpyEnvironment(options))
else:
objEnv = ElecpyEnvironment(options)
objEnv.env_init()
for epi in range(3):
print 'Episode {0}'.format(epi)
objEnv.env_start()
cnt_step = 0
while True:
cnt_step += 1
action = Action(numInts=1)
action.intArray = [0]
rot = objEnv.env_step(action)
if rot.terminal:
break
base = self.var[0] if self.pos[1] <= 1.0 else self.var[1]
a = self.var[2]
return base - (int(self.pos[0]) % 5) * a
elif intAction < 4:
return -1.0
elif intAction >= 4:
return -1.4
else:
print "ERROR in FuelWorld.takeAction"
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser(
description=
'Run 2D MultiRoom Noisy Continuous Gridworld environment in network mode.'
)
gridworld.addGridworldArgs(parser)
parser.add_argument(
"--fuel_noise",
type=float,
default=0.0,
help=
"If non-zero then gives the standard deviation of the additive Gaussian noise to add to the fuel expenditure."
)
args = parser.parse_args()
EnvironmentLoader.loadEnvironment(
FuelWorld(noise=args.noise,
fudge=args.fudge,
fuel_noise=args.fuel_noise))
theReward=-1
episodeOver=1
if self.currentState >= 20:
self.currentState=20
theReward=1
episodeOver=1
theObs=Observation()
theObs.intArray=[self.currentState]
returnRO=Reward_observation_terminal()
returnRO.r=theReward
returnRO.o=theObs
returnRO.terminal=episodeOver
return returnRO
def env_cleanup(self):
pass
def env_message(self,inMessage):
if inMessage=="what is your name?":
return "my name is skeleton_environment, Python edition!";
else:
return "I don't know how to respond to your message";
if __name__=="__main__":
EnvironmentLoader.loadEnvironment(skeleton_environment())
returnRO.terminal = episodeOver
return returnRO
def env_cleanup(self):
pass
def env_message(self, inMessage):
return "I don't know how to respond to your message"
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser(
description='Run Noisy Acrobot environment in network mode.')
parser.add_argument(
"--noise",
type=float,
default=0,
help=
"Standard deviation of additive noise to generate, affects the action effects."
)
parser.add_argument("--random_restarts",
type=bool,
default=False,
help="Restart the state with random values.")
args = parser.parse_args()
EnvironmentLoader.loadEnvironment(
Acrobot(noise=args.noise, random_start=args.random_restarts))
pygame.display.flip()
pygame.quit()
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Pinball domain')
parser.add_argument('configuration', help='the configuration file')
parser.add_argument('--width',
action='store',
type=int,
default=500,
help='screen width (default: 500)')
parser.add_argument('--height',
action='store',
type=int,
default=500,
help='screen height (default: 500)')
parser.add_argument('-r',
'--rlglue',
action='store_true',
help='expose the environment through RL-Glue')
args = parser.parse_args()
if args.rlglue:
print 'Starting rl-glue'
EnvironmentLoader.loadEnvironment(PinballRLGlue(args.configuration))
else:
run_pinballview(args.width, args.height, args.configuration)
terminal = 0
if self.stepCount == 200:
terminal = 1
ro = Reward_observation_terminal()
ro.r = 1.0
ro.o = self.o
ro.terminal = terminal
return ro
self.o.intArray = range(0, 5)
#cheating, might break something
self.o.doubleArray = range(0, 5)
terminal = 0
if self.stepCount == 5000:
terminal = 1
ro = Reward_observation_terminal()
ro.r = 1.0
ro.o = self.o
ro.terminal = terminal
return ro
def env_cleanup(self):
pass
def env_message(self, inMessage):
return None
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(test_speed_environment())
returnRO.terminal = int(self.counter >= self.T)
return returnRO
def env_cleanup(self):
pass
def env_message(self,inMessage):
return "I don't know how to respond to your message";
if __name__=="__main__":
import argparse
parser = argparse.ArgumentParser(description='Run 2D Noisy Continuous Gridworld environment in network mode.')
parser.add_argument("--demand_mean", type=float, default=10., help="Mean demand for the product.")
parser.add_argument("--demand_std", type=float, default=1.,
help="Standard deviation of demand for the product.")
parser.add_argument("--payoff", type=float, default=5., help="Payment received per unit product sold.")
parser.add_argument("--cost", type=float, default=2., help="Cost per unit product purchased.")
parser.add_argument("--discount_factor", type=float, default=0.999, help="Discount factor to learn over.")
parser.add_argument("--time_period", type=int, default=20, help="Time period for problem. (Number of steps to run)")
parser.add_argument("--noise", type=float, default=0, help="Standard deviation of additive noise to generate")
args = parser.parse_args()
EnvironmentLoader.loadEnvironment(BatchReplenishment(demand_mean=args.demand_mean,
demand_std=args.demand_std,
payoff=args.payoff,
cost=args.cost,
gamma=args.discount_factor,
time_period = args.time_period,
noise=args.noise))
rot.o = observation
current_map = 'map\n'
for i in range(0, len(self.map), self.n_cols):
current_map += ' '.join(map(str,
self.map[i:i + self.n_cols])) + '\n'
if (i % 16 == 0):
current_map += "\n"
self.history.append(current_map)
if rot.r == -1:
f = open('history.txt', 'a')
history = '\n'.join(self.history)
f.writelines('# START\n' + history + '# END\n\n')
f.close()
# 決着がついた場合は agentのagent_end
# 決着がついていない場合は agentのagent_step に続く
return rot
def env_cleanup(self):
pass
def env_message(self, message):
pass
if __name__ == '__main__':
EnvironmentLoader.loadEnvironment(MarubatsuEnvironment())
"""
Random player environment
"""
import random
from rlglue.environment import EnvironmentLoader
from wrapper_environment import WrapperEnvironment
class RandomEnvironment(WrapperEnvironment):
name = 'random'
def env_play(self):
"""
Pick the first free spot, and play there.
"""
open_spots = []
for i in range(len(self.state)):
if self.state[i] == 0:
open_spots.append(i)
self.state[random.choice(open_spots)] = self.color
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(RandomEnvironment())
1, 1, 2, 2, 2, 0, 2, 2, 2, 0, 0, 0, 0, 0, 0, 2, 2, 2, 0, 0, 0, 0,
0, 0, 2, 2, 2, 0, 0, 0, 0, 0, 0, 2, 2, 2, 0, 0, 0, 0, 0, 0
]
numRows = len(self.worldmap)
numCols = len(self.worldmap[0])
for row in range(0, numRows):
print
print "Row: " + str(row) + " ",
for col in range(0, numCols):
if self.checkValid(row, col):
flat = self.calculateFlatState(row, col)
flat_i = self.validstates.index(flat)
print unicode(argmaxes[flat_i]),
else:
print "X",
"""
if self.map[row][col] == self.GOAL:
print "G",
if self.map[row][col] == self.WALL:
print "X",
if self.map[row][col] == self.START:
print "S",
if self.map[row][col] == self.FREE:
print " ",
"""
print
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(threeroom_environment())
if inMessage.startswith("set-start-state"):
splitString = inMessage.split(" ")
self.state = array(eval(splitString[1]))
self.fixedStartState = True
return "Message understood. Using fixed start state."
return "InvasiveEnvironment(Python) does not respond to messages."
def setAgentState(self, S):
assert len(S)==self.simulationParameterObj.habitatSize*self.simulationParameterObj.nbrReaches
self.state = S
valid = True
return valid
def setRandomState(self):
S = array([random.randint(1, 3) for i in
xrange(self.simulationParameterObj.nbrReaches * self.simulationParameterObj.habitatSize)])
self.setAgentState(S)
def checkValid(self, S):
valid = True
return valid
def printState(self):
print "Agent is at: " + str(self.state)
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(
InvasiveEnvironment(simulationParameterObj=None, actionParameterObj=None, Bad_Action_Penalty=-10000,fixedStartState=False, nbrReaches=7,
habitatSize=4, seed=1))
def printState(self):
numRows = len(self.map)
numCols = len(self.map[0])
print "Agent is at: " + str(self.agentRow) + "," + str(self.agentCol)
print "Columns:0-10 10-17"
print "Col ",
for col in range(0, numCols):
print col % 10,
for row in range(0, numRows):
print
print "Row: " + str(row) + " ",
for col in range(0, numCols):
if self.agentRow == row and self.agentCol == col:
print "A",
else:
if self.map[row][col] == self.WORLD_GOAL:
print "G",
if self.map[row][col] == self.WORLD_MINE:
print "M",
if self.map[row][col] == self.WORLD_OBSTACLE:
print "*",
if self.map[row][col] == self.WORLD_FREE:
print " ",
print
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(mines_environment())
# (string) -> string
def env_message(self, inMessage):
if inMessage.startswith("print"):
self.toprint = 1
return "message understood, print"
if inMessage.startswith("stop print"):
self.toprint = 0
return "message understood, stop print"
return "RmaxAgent(Python) does not understand your message."
def clearscreen(self, numlines=100):
"""Clear the console.
numlines is an optional argument used only as a fall-back.
"""
# Thanks to Steven D'Aprano, http://www.velocityreviews.com/forums
if os.name == "posix":
# Unix/Linux/MacOS/BSD/etc
os.system('clear')
elif os.name in ("nt", "dos", "ce"):
# DOS/Windows
os.system('CLS')
else:
# Fallback for other operating systems.
print('\n' * numlines)
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(My_Environment())
if msg.startswith('set'):
param, value = msg.split(None, 2)[1:]
self.debug('set', param, value)
self.env_message_set_param(param, value)
elif msg.startswith('get'):
param = msg.split(None, 1)[1]
return self.env_message_get_param(param)
else:
return self.env_message_handler(msg)
def debug(self, *args):
""" Print a debug msg """
if self.debug:
args = [str(a) for a in args]
print "%s: %s" % (self.name, ' '.join(args))
def step_out(self, *args):
if self.output_steps:
args = [str(a) for a in args]
print ' '.join(args)
if __name__ == '__main__':
#p = PuddleEnvironment()
#p.env_start()
EnvironmentLoader.loadEnvironment(PuddleEnvironment())
self.nonEmptyObservation.charArray=['a','b','c','d','e'] return "" def env_start(self): self.whichEpisode=self.whichEpisode+1 if self.whichEpisode % 2 == 0: return self.emptyObservation else: return self.nonEmptyObservation def env_step(self,action): ro=Reward_observation_terminal() if self.whichEpisode % 2 == 0: ro.o=self.emptyObservation else: ro.o=self.nonEmptyObservation return ro def env_cleanup(self): pass def env_message(self,inMessage): return None if __name__=="__main__": EnvironmentLoader.loadEnvironment(test_empty_environment())
return returnRO
def env_cleanup(self):
pass
def env_message(self,inMessage):
return "I don't know how to respond to your message";
if __name__=="__main__":
import argparse
path_to_problems = os.path.join(os.path.dirname(__file__), 'configs', 'pomdps', '*')
config_files = glob.glob(path_to_problems)
parser = argparse.ArgumentParser(description='Run a POMDP problem file as a domain in RL-Glue in network mode.')
group = parser.add_mutually_exclusive_group(required=True)
group.add_argument("--file", type=str, default=config_files[0],
help="Run POMDP domain given the path to a POMDP problem file.")
group.add_argument("--list", action='store_true', default=False, help="List path to included POMDP problem files.")
args = parser.parse_args()
if args.list:
print "Included POMDP problem files:"
for file in config_files:
print file
else:
EnvironmentLoader.loadEnvironment(POMDPEnvironment(spec_filename=args.file))
parser = argparse.ArgumentParser(description='Run a specified POMDP in RL-Glue in network mode.')
parser.add_argument("--pomdp_file", type=str, help="Filename for POMDP spec file to load and use.", required=True)
args = parser.parse_args()
EnvironmentLoader.loadEnvironment(POMDPEnvironment(args.pomdp_file))
#print("newState: "+str(self.state))
r = 0.0
if np.any(self.state_ranges[:,0] > self.state[:]) or \
np.any(self.state_ranges[:,1] < self.state[:]):
# r = -1
r = -np.sum(3.0 * self.state_ranges[:,1]**2)
r *= 6000-self.num_sim_steps
terminate = True
else:
# perr = np.linalg.norm(self.prevState[:2] - self.state_goal[:2])
# nerr = np.linalg.norm(self.state[:2] - self.state_goal[:2])
# r = math.exp(-np.sum(abs(self.state[:2]-self.state_goal[:2])/(.1*(self.state_ranges[:2,1]-self.state_ranges[:2,0]))))* \
# math.exp(-np.sum(abs(self.state[3:5]-self.state_goal[3:5])/(.1*(self.state_ranges[3:5,1]-self.state_ranges[3:5,0]))))* \
# math.exp(-np.sum(abs(self.state[6:8]-self.state_goal[6:8])/(.1*(self.state_ranges[6:8,1]-self.state_ranges[6:8,0]))))
# r = math.exp(-np.sum(abs(self.state[:2]-self.state_goal[:2])/(.1*(self.state_ranges[:2,1]-self.state_ranges[:2,0]))))* \
# math.exp(-np.sum(abs(self.state[3:5]-self.state_goal[3:5])/(.1*(self.state_ranges[3:5,1]-self.state_ranges[3:5,0]))))
# r -= (np.sum(((self.state[:2]-self.state_goal[:2])/(self.state_ranges[:2,1]-self.state_ranges[:2,0]))**2)+ \
# np.sum(((self.state[3:5]-self.state_goal[3:5])/(self.state_ranges[3:5,1]-self.state_ranges[3:5,0]))**2))
r -= (self.state[0]-self.state_goal[0])**2
r -= (self.state[1]-self.state_goal[1])**2
r -= self.state[3]**2
r -= self.state[4]**2
terminate = False
print("reward "+str(r))
return r,terminate
if __name__=="__main__":
EnvironmentLoader.loadEnvironment(vrep_environment())
self.fixedStartState = True
return "Message understood. Using fixed start state."
return "InvasiveEnvironment(Python) does not respond to messages."
def setAgentState(self, S):
assert len(S)==self.simulationParameterObj.habitatSize*self.simulationParameterObj.nbrReaches
self.state = S
valid = True
return valid
def setRandomState(self):
S = array([random.randint(1, 3) for i in
xrange(self.simulationParameterObj.nbrReaches * self.simulationParameterObj.habitatSize)])
self.setAgentState(S)
def checkValid(self, S):
valid = True
return valid
def printState(self):
print "Agent is at: " + str(self.state)
# ============================ PARAMETERS =====================================
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(
InvasiveEnvironment(simulationParameterObj=None, actionParameterObj=None, Bad_Action_Penalty=-10000,fixedStartState=False, nbrReaches=7,
habitatSize=4, seed=1))
# ============================ PARAMETERS =====================================
#return max(outcomes)
max_element = -1
for o in outcomes:
if o == +1:
return o
max_element = max(o,max_element)
return max_element
finally:
board.undoMove(move)
moves = [(move, evaluateMove(move)) for move in board.getValidMoves()]
random.shuffle(moves)
moves.sort(key = lambda (move, winner): winner)
board.makeMove(moves[-1][0], player)
class MiniMaxEnvironment(WrapperEnvironment):
name = 'minimax'
def env_play(self):
b = Board(self.state)
computerPlayer(b, Player_X)
b.output()
self.state = b.pieces
if __name__ == "__main__":
#game()
EnvironmentLoader.loadEnvironment(MiniMaxEnvironment())
self.nonEmptyObservation.charArray = ['a', 'b', 'c', 'd', 'e']
return ""
def env_start(self):
self.whichEpisode = self.whichEpisode + 1
if self.whichEpisode % 2 == 0:
return self.emptyObservation
else:
return self.nonEmptyObservation
def env_step(self, action):
ro = Reward_observation_terminal()
if self.whichEpisode % 2 == 0:
ro.o = self.emptyObservation
else:
ro.o = self.nonEmptyObservation
return ro
def env_cleanup(self):
pass
def env_message(self, inMessage):
return None
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(test_empty_environment())
"""
Environment which always plays the first free spot available on the board.
"""
from rlglue.environment import EnvironmentLoader
from wrapper_environment import WrapperEnvironment
class FirstFreeEnvironment(WrapperEnvironment):
name = 'first_free'
def env_play(self):
"""
Pick the first free spot, and play there.
"""
for i in range(len(self.state)):
if self.state[i] == 0:
self.state[i] = self.color
return
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(FirstFreeEnvironment())
if event.type == pygame.QUIT:
done = True
if event.type == pygame.KEYUP or event.type == pygame.KEYDOWN:
user_action = actions.get(event.key, PinballModel.ACC_NONE)
if environment.take_action(user_action) == environment.END_EPISODE:
done = True
environment_view.blit()
pygame.display.flip()
pygame.quit()
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Pinball domain')
parser.add_argument('configuration', help='the configuration file')
parser.add_argument('--width', action='store', type=int,
default=500, help='screen width (default: 500)')
parser.add_argument('--height', action='store', type=int,
default=500, help='screen height (default: 500)')
parser.add_argument('-r', '--rlglue', action='store_true', help='expose the environment through RL-Glue')
args = parser.parse_args()
if args.rlglue:
print 'Starting rl-glue'
EnvironmentLoader.loadEnvironment(PinballRLGlue(args.configuration))
else:
run_pinballview(args.width, args.height, args.configuration)
def main(): EnvironmentLoader.loadEnvironment(SumoEnvironment(sys.argv[1]))
returnObs.intArray=np.append(np.zeros(128), [ item for innerlist in arr for item in innerlist ])
scipy.misc.imsave('screen.png', arr)
returnRO=Reward_observation_terminal()
returnRO.r=theReward
returnRO.o=returnObs
returnRO.terminal=episodeOver
# イベント処理
for event in pygame.event.get():
if event.type == QUIT: # 終了イベント
sys.exit()
return returnRO
def env_cleanup(self):
pass
def env_message(self,inMessage):
if inMessage=="what is your name?":
return "my name is dqn_environment, Python edition!";
else:
return "I don't know how to respond to your message";
if __name__ == "__main__":
App = EdgeTracer()
#App.run()
EnvironmentLoader.loadEnvironment(EdgeTracer())
print('main start')
if self.inFuelCell(self.pos):
self.fuel += 20.0
if self.fuel > 60.0:
self.fuel = 60.0
if gridworld.Gridworld.isAtGoal(self):
return 0.0
elif self.fuel < 0:
return -400.0
elif self.inFuelCell(self.pos): # Fuel costs
base = self.var[0] if self.pos[1] <= 1.0 else self.var[1]
a = self.var[2]
return base - (int(self.pos[0]) % 5)*a
elif intAction < 4:
return -1.0
elif intAction >= 4:
return -1.4
else:
print "ERROR in FuelWorld.takeAction"
if __name__=="__main__":
import argparse
parser = argparse.ArgumentParser(description='Run 2D MultiRoom Noisy Continuous Gridworld environment in network mode.')
gridworld.addGridworldArgs(parser)
parser.add_argument("--fuel_noise", type=float, default=0.0,
help="If non-zero then gives the standard deviation of the additive Gaussian noise to add to the fuel expenditure.")
args = parser.parse_args()
EnvironmentLoader.loadEnvironment(FuelWorld(noise=args.noise, fudge=args.fudge, fuel_noise=args.fuel_noise))
"the puddle's depth.")
parser.add_argument("--puddle_penalty", type=float, default=-100,
help="The reward penalty scale for walking through puddles.")
args = parser.parse_args()
kwargs = {}
if args.puddle is not None:
means = []
covs = []
for puddle in args.puddle:
means.append(tuple(puddle[:2]))
covs.append(tuple(puddle[2:]))
kwargs['puddle_means'] = means
kwargs['puddle_var'] = covs
if args.size_x:
kwargs['size_x'] = args.size_x
if args.size_y:
kwargs['size_y'] = args.size_y
if args.goal_x:
kwargs['goal_x'] = args.goal_x
if args.goal_y:
kwargs['goal_y'] = args.goal_y
if args.noise:
kwargs['noise'] = args.noise
if args.fudge:
kwargs['fudge'] = args.fudge
if args.random_restarts:
kwargs['random_start'] = args.random_restarts
EnvironmentLoader.loadEnvironment(PuddleWorld(**kwargs))
'-t','--test',
dest='test', action='store_true',default=False,
help="test mode")
parser.add_option(
'-p','--param_file',
dest='param_file', action='store', type='string', default='./rlglue_param.json',
help="json file of simulation parameters")
parser.add_option(
'-d','--dst',
dest='savepath', action='store', type='string', default='./result/data',
help="Save data path.")
(options, args) = parser.parse_args()
#print 'options', options
if not options.test:
EnvironmentLoader.loadEnvironment(ElecpyEnvironment(options))
else:
objEnv = ElecpyEnvironment(options)
objEnv.env_init()
for epi in range(3):
print 'Episode {0}'.format(epi)
objEnv.env_start()
cnt_step = 0
while True:
cnt_step += 1
action = Action(numInts=1)
action.intArray = [0]
rot = objEnv.env_step(action)
if rot.terminal:
break
def env_init(self): return "" def env_start(self): return Observation() def env_step(self,action): return Reward_observation_terminal() def env_cleanup(self): pass def env_message(self,inMessage): if inMessage==None: return "null" if inMessage=="": return "empty" if inMessage=="null": return None if inMessage=="empty": return "" return inMessage; if __name__=="__main__": EnvironmentLoader.loadEnvironment(test_message_environment())
theReward = -1
episodeOver = 1
if self.currentState >= 20:
self.currentState = 20
theReward = 1
episodeOver = 1
theObs = Observation()
theObs.intArray = [self.currentState]
returnRO = Reward_observation_terminal()
returnRO.r = theReward
returnRO.o = theObs
returnRO.terminal = episodeOver
return returnRO
def env_cleanup(self):
pass
def env_message(self, inMessage):
if inMessage == "what is your name?":
return "my name is skeleton_environment, Python edition!"
else:
return "I don't know how to respond to your message"
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(skeleton_environment())
parser = argparse.ArgumentParser(
description=
'Run Noisy Cart Pole Balancing or Swing Up environment in network mode.'
)
parser.add_argument(
"--noise",
type=float,
default=0,
help=
"Standard deviation of additive noise to generate, affects the action effects."
)
parser.add_argument(
"--random_restarts",
type=bool,
default=False,
help="Restart the cart with a random location and velocity.")
parser.add_argument(
"--mode",
choices=["easy", "hard", "swingup"],
default="easy",
type=str,
help=
"Choose the type of cart pole domain. Easy/hard balancing, or swing up."
)
args = parser.parse_args()
EnvironmentLoader.loadEnvironment(
CartPole(mode=args.mode,
noise=args.noise,
random_start=args.random_restarts))
type=float,
default=-100,
help="The reward penalty scale for walking through puddles.")
args = parser.parse_args()
kwargs = {}
if args.puddle is not None:
means = []
covs = []
for puddle in args.puddle:
means.append(tuple(puddle[:2]))
covs.append(tuple(puddle[2:]))
kwargs['puddle_means'] = means
kwargs['puddle_var'] = covs
if args.size_x:
kwargs['size_x'] = args.size_x
if args.size_y:
kwargs['size_y'] = args.size_y
if args.goal_x:
kwargs['goal_x'] = args.goal_x
if args.goal_y:
kwargs['goal_y'] = args.goal_y
if args.noise:
kwargs['noise'] = args.noise
if args.fudge:
kwargs['fudge'] = args.fudge
if args.random_restarts:
kwargs['random_start'] = args.random_restarts
EnvironmentLoader.loadEnvironment(PuddleWorld(**kwargs))
ro.r=1.0 else: self.o.doubleArray=[0.0078125,-0.0078125,0.0,0.0078125e150,-0.0078125e150] self.o.charArray=['g','F','?',' ','&'] self.o.intArray=[173,-173,2147483647,0,-2147483648] ro.r=-2.0 ro.o=self.o ro.terminal=terminal return ro def env_cleanup(self): pass def env_message(self,inMessage): timesToPrint=self.stepCount%3 outMessage=inMessage+"|" for i in range(0, timesToPrint): outMessage=outMessage+"%d" % (self.stepCount) outMessage=outMessage+"." outMessage=outMessage+"|"+inMessage return outMessage if __name__=="__main__": EnvironmentLoader.loadEnvironment(test_1_environment())
if __name__ == "__main__":
pygame.init()
pygame.freetype.init()
gameArea = pygame.Rect([0, 0], size)
if len(sys.argv) > 1:
useGlue = (sys.argv[1] == 'True')
else:
useGlue = False
black = 0, 0, 0
screen = pygame.display.set_mode([size[0],
int(size[1] * (4 / 3.0))
]) #,pygame.FULLSCREEN)
count = 0
if len(sys.argv) > 2:
p1Type = int(sys.argv[2])
else:
p1Type = 0
if len(sys.argv) > 3:
p2Type = int(sys.argv[3])
else:
p2Type = 1
if useGlue:
EnvironmentLoader.loadEnvironment(World(p1Type, 2))
else:
world = World(p1Type, p2Type)
world.start()
while True or count < 300:
world.step()
count += 1
def env_init(self):
return ""
def env_start(self):
return Observation()
def env_step(self, action):
return Reward_observation_terminal()
def env_cleanup(self):
pass
def env_message(self, inMessage):
if inMessage == None:
return "null"
if inMessage == "":
return "empty"
if inMessage == "null":
return None
if inMessage == "empty":
return ""
return inMessage
if __name__ == "__main__":
EnvironmentLoader.loadEnvironment(test_message_environment())
def __init__(self, size_x=10, size_y=10, goal_x=10, goal_y=10, wind_center=7., wind_stdev=1.0, wind_power=2.0, noise=0.0, random_start=False, fudge=1.4143):
gridworld.Gridworld.__init__(self, size_x=size_x, size_y=size_y, goal_x=goal_x, goal_y=goal_y, noise=noise, random_start=random_start, fudge=fudge)
self.wind_center = wind_center
self.wind_stdev = wind_stdev
self.wind_power = wind_power
self.domain_name = "Continuous Windy Gridworld by Will Dabney"
def reset(self):
if self.random_start:
self.pos = numpy.random.random((2,)) * self.size
else:
self.pos = numpy.array([0.0, self.size[1]*0.5])
def takeAction(self, action):
self.pos[1] += norm.pdf(self.pos[0], self.wind_center, self.wind_stdev) * self.wind_power
return gridworld.Gridworld.takeAction(self, action)
if __name__=="__main__":
import argparse
parser = argparse.ArgumentParser(description='Run 2D MultiRoom Noisy Continuous Gridworld environment in network mode.')
gridworld.addGridworldArgs(parser)
parser.add_argument("--wind_center", type=float, default=7, help="Center, or strongest point, in the x-direction of the wind")
parser.add_argument("--wind_scale", type=float, default=1.0, help="Scale, or width, of the wind effects around the center.")
parser.add_argument("--wind_power", type=float, default=2.0, help="The power, or strength, of the wind.")
args = parser.parse_args()
EnvironmentLoader.loadEnvironment(
WindyGridworld(args.size_x, args.size_y, args.goal_x, args.goal_y, wind_center=args.wind_center,
wind_stdev=args.wind_scale, wind_power=args.wind_power, noise=args.noise,
random_start=args.random_restarts, fudge=args.fudge))