def __init__(self, env, gamma=.99):
grid = EnvMDP.to_grid_matrix(env)
reward = {}
states = set()
self.rows = len(grid)
self.cols = len(grid[0])
self.grid = grid
for x in range(self.cols):
for y in range(self.rows):
if grid[y][x] is not None:
states.add((x, y))
reward[(x, y)] = grid[y][x]
self.states = states
terminals = EnvMDP.to_position(env, letter=b'GH')
actlist = list(range(env.action_space.n))
transitions = EnvMDP.to_transitions(env)
init = EnvMDP.to_position(env, letter=b'S')[0]
MDP.__init__(self,
init,
actlist=actlist,
terminals=terminals,
transitions=transitions,
reward=reward,
states=states,
gamma=gamma)
def __init__(self, grid, goalVals, discount=.99, tau=.01, epsilon=.001): MDP.__init__(self, discount=discount, tau=tau, epsilon=epsilon) self.goalVals = goalVals self.grid = grid self.setGridWorld() self.valueIteration() self.extractPolicy()
def __init__(self, grid, goalVals, discount=.99, tau=.01, epsilon=.001):
MDP.__init__(self, discount=discount, tau=tau, epsilon=epsilon)
self.goalVals = goalVals
self.grid = grid
self.setGridWorld()
self.valueIteration()
self.extractPolicy()
def __init__(self, desc=None, map_name="4x4", slip_chance=0.2):
if desc is None and map_name is None:
raise ValueError('Must provide either desc or map_name')
elif desc is None:
desc = self.MAPS[map_name]
assert ''.join(desc).count('S') == 1, "this implementation supports having exactly one initial state"
assert all(c in "SFHG" for c in ''.join(desc)), "all cells must be either of S, F, H or G"
self.desc = desc = np.asarray(list(map(list,desc)),dtype='str')
self.lastaction = None
nrow, ncol = desc.shape
states = [(i, j) for i in range(nrow) for j in range(ncol)]
actions = ["left","down","right","up"]
initial_state = states[np.array(desc == b'S').ravel().argmax()]
def move(row, col, movement):
if movement== 'left':
col = max(col-1,0)
elif movement== 'down':
row = min(row+1,nrow-1)
elif movement== 'right':
col = min(col+1,ncol-1)
elif movement== 'up':
row = max(row-1,0)
else:
raise("invalid action")
return (row, col)
transition_probs = {s : {} for s in states}
rewards = {s : {} for s in states}
for (row,col) in states:
if desc[row, col] in "GH": continue
for action_i in range(len(actions)):
action = actions[action_i]
transition_probs[(row, col)][action] = {}
rewards[(row, col)][action] = {}
for movement_i in [(action_i - 1) % len(actions), action_i, (action_i + 1) % len(actions)]:
movement = actions[movement_i]
newrow, newcol = move(row, col, movement)
prob = (1. - slip_chance) if movement == action else (slip_chance / 2.)
if prob == 0: continue
if (newrow, newcol) not in transition_probs[row,col][action]:
transition_probs[row,col][action][newrow, newcol] = prob
else:
transition_probs[row, col][action][newrow, newcol] += prob
if desc[newrow, newcol] == 'G':
rewards[row,col][action][newrow, newcol] = 1.0
MDP.__init__(self, transition_probs, rewards, initial_state)
def __init__(self,
rows,
cols,
definitiveness,
initstate,
terminals,
obstacles,
gamma=.9):
self.rows = rows
self.cols = cols
self.definitiveness = definitiveness
self.initstate = initstate
self.terminals = terminals
self.obstacles = obstacles
stateset = set()
for y in range(1, self.cols + 1):
for x in range(1, self.rows + 1):
stateset.add((x, y))
actionset = {'up', 'down', 'right', 'left'}
MDP.__init__(self, stateset, actionset, gamma)
