taskToPerform = args.task
epsilon = args.epsilon
verbose = args.verbose
inputMDP = args.input
outputPath = args.output
optionsToLoad = args.load
bothDirections = args.both
num_seeds = args.num_seeds
max_length_episode = args.max_length_ep
num_episodes = args.num_episodes
if not verbose:
warnings.filterwarnings('ignore')
# Create environment
env = GridWorld(path=inputMDP, useNegativeRewards=False)
numStates = env.getNumStates()
numRows, numCols = env.getGridDimensions()
# I may load options if I'm told so:
loadedOptions = None
if optionsToLoad != None:
loadedOptions = []
for i in xrange(len(optionsToLoad)):
loadedOptions.append(Utils.loadOption(optionsToLoad[i]))
plot = Plotter(outputPath, env)
plot.plotPolicy(loadedOptions[i], str(i + 1) + '_')
if taskToPerform == 1: #Discover options
optionDiscoveryThroughPVFs(env=env,
epsilon=epsilon,
# [' ','#',' ',' ',' '],
# [' ','#', 1,'#', 10],
# ['S',' ',' ',' ',' '],
# [-10,-10, -10, -10, -10]]
# grid = [[' ',' ',' ',+1],
# ['#','#',' ','#'],
# [' ','#',' ',' '],
# [' ','#','#',' '],
# ['S',' ',' ',' ']]
grid = [[ '#',-100, -100, -100, -100, -100, '#'],
[ 1, 'S', ' ', ' ', ' ', ' ', 10],
[ '#',-100, -100, -100, -100, -100, '#']]
Grid = GridWorld(grid, 0.01, -1.0)
M = MarkovDecisonProcess(0.4, Grid, 100)
M.ValueIteration()
values = M.getValues()
actions = M.getActions()
qvalues = M.getQValues()
for i in values.keys():
print(f"{i} : {values[i]}, {actions[i]}")
RL = QLearning(Grid, 50000, alpha = 0.6, epsilon = 0.2, discount = 0.4)
RL.train()
values = RL.getValues()
actions = RL.getActions()
qvalues = RL.getQValues()
for i in values.keys():
print(f"{i} : {values[i]}, {actions[i]}")