def alphabeta_dtm(sp, a, s, depth, alpha, beta):
"""
alpha beta pruning on a ground truth dtm
params:
sp: previous state
a: action
s: current state
depth: depth of search
alpha
beta
returns:
max_a: best action
max_score: score of this action
"""
if depth == 0:
ep = Environment(sp)
return None, -map_side_to_int(ep.get_turn()) * ep.action_outcome(a)
elif depth == 1:
e = Environment(s)
as_pairs = e.get_as_pairs()
outcomes = [
-0.5 * map_side_to_int(e.get_turn()) * e.action_outcome(an)
for (an, sn) in as_pairs
]
max_o = max(outcomes)
rand = outcomes.index(max_o)
return as_pairs[rand][0], max_o
else:
best_an = None
e = Environment(s)
as_pairs = e.get_as_pairs()
for (an, sn) in as_pairs:
score = -0.5 * alphabeta_dtm(s, an, sn, depth - 1, -beta,
-alpha)[1]
if score >= beta:
return an, beta
elif score > alpha:
alpha = score
best_an = an
return best_an, alpha
def alphabeta_outcome(sp, a, s, depth, alpha, beta):
"""
alpha beta pruning on a ground truth outcome
params:
sp: previous state
a: action
s: current state
depth: depth of search
alpha
beta
returns:
max_a: best action
max_score: score of this action
"""
if depth < 1:
env = Environment(sp)
env.perform_action(a)
o = map_side_to_int(env.get_turn()) * env.int_outcome()
#print o
return None, o
env = Environment(s)
as_pairs = env.get_as_pairs()
if len(as_pairs) == 0:
env = Environment(sp)
env.perform_action(a)
o = map_side_to_int(env.get_turn()) * env.int_outcome()
#print o
return None, o
if depth == 1:
outcomes = [
0.5 * map_side_to_int(env.get_turn()) * env.action_outcome(a)
for (a, sn) in as_pairs
]
best = np.argmax(np.array(outcomes))
best_o = outcomes[best]
return as_pairs[best][0], best_o
act = None
for (a, sn) in as_pairs:
score = -0.5 * alphabeta_outcome(s, a, sn, depth - 1, -beta, -alpha)[1]
if score >= beta:
return a, beta
elif score > alpha:
alpha = score
act = a
return act, alpha
