def max_reward(self, state, action):
new_state = move(state, action)
if not new_state in self.R.keys():
self.R[new_state] = []
for action in self.actions:
self.R[new_state].append(self.reward(new_state, action))
return max(self.R[new_state])
def generate_next_states(puzzle):
next_states = []
moves = find_possible_moves(puzzle)
for m in moves:
state = move(puzzle, m)
next_states.append(state)
return next_states
def makeNearGoal():
cube = State()
cube.set_top([['W', 'W', 'W'], ['W', 'W', 'W'], ['W', 'W', 'W']])
cube.set_bottom([['B', 'B', 'B'], ['B', 'B', 'B'], ['B', 'B', 'B']])
cube.set_left([['O', 'O', 'O'], ['G', 'G', 'G'], ['G', 'G', 'G']])
cube.set_right([['G', 'G', 'G'], ['O', 'O', 'O'], ['O', 'O', 'O']])
cube.set_front([['R', 'R', 'R'], ['Y', 'Y', 'Y'], ['Y', 'Y', 'Y']])
cube.set_back([['Y', 'Y', 'Y'], ['R', 'R', 'R'], ['R', 'R', 'R']])
for action in cube.actions:
new_s = move(cube, action)
print(action)
if new_s.isGoalState():
print("executing the " + action +
" action resulted in the below goal state " + str(new_s))
def reward(self, state, action):
# this reward function should be a function approximation made up of
# a set of features, these features should be in decreasing order of priority:
# 1. solved sides ()
# use next state to get value for next state vs. self.curr_state, to determine
# if feature values should be 1 or 0, e.g. if solved_sides(next_state) > solved_sides(self.curr_state)
# then the solved sides feature is 1, else 0
next_state = move(state, action)
if next_state.isGoalState():
print(state)
print(next_state)
print("REWARD IS GOAL")
return 100
reward = -0.1
solved_sides = 2 * (num_solved_sides(next_state) < num_solved_sides(state))
solved_pieces = 0.5 * (num_pieces_correct_side(next_state) < num_pieces_correct_side(state))
if (next_state.__hash__(), action) in self.QV.keys():
reward -= 0.2
reward -= solved_sides
reward -= solved_pieces
return reward
def register_patterns(self):
s = State()
# get list of goal successors
for action in self.actions:
s_ = move(s, action)
self.one_away.append(s_)
for action_ in self.actions:
self.QV[(s_.__hash__(), action_)] = -10 if action_ != action else 10
# get list of successors of goal successors
for s in self.one_away:
for action in self.actions:
s_ = move(s, action)
self.two_away.append(s_)
for action_ in self.actions:
self.QV[(s_.__hash__(), action_)] = -6 if action_ != action else 6
# get list of successors of successors of goal successors
for s in self.two_away:
for action in self.actions:
s_ = move(s, action)
self.three_away.append(s_)
for action_ in self.actions:
self.QV[(s_.__hash__(), action_)] = -5 if action_ != action else 5
for s in self.three_away:
for action in self.actions:
s_ = move(s, action)
self.four_away.append(s_)
for action_ in self.actions:
self.QV[(s_.__hash__(), action_)] = -4 if action_ != action else 4
for s in self.four_away:
for action in self.actions:
s_ = move(s, action)
self.five_away.append(s_)
for action_ in self.actions:
self.QV[(s_.__hash__(), action_)] = -3 if action_ != action else 3
for s in self.five_away:
for action in self.actions:
s_ = move(s, action)
self.six_away.append(s_)
for action_ in self.actions:
self.QV[(s_.__hash__(), action_)] = -1 if action_ != action else 1