def parse_binary_search(sent_len, scorer, m):
def binary_search(seq, t):
min = -10
max = 10
for i in range(25):
if max < min:
return -1
m = (min + max) / 2.0
size = seq(m)
#print t, m, size
if size < t:
min = m
elif size > t:
max = m
else:
return m
return m
c = Chart(sent_len+1)
interface.Parser().parse_bigram(sent_len, scorer, None, c)
counts = c.counts()
# hypergraph = c.hypergraph
# pot = c.pot
# def score(label, tail_size):
# typ, d, s, t, _ = label
# if typ == interface.Trap:
# if d == interface.Left: s, t = t, s
# return scorer.arc_score(s, t)
# if typ == interface.Tri and d == interface.Right and tail_size == 1:
# return scorer.bigram_score(s, t+1) - \
# ((t+1 - s - 1) * scorer.skip_penalty) if s != t+1 else 0.0
# return 0.0
def f(pen):
scorer.skip_penalty = pen
# c.score = score
parse = interface.make_parse(sent_len+1, c.regen(pen, counts))
return sent_len - parse.skipped_words()
pen = binary_search(f, m)
scorer.skip_penalty = pen
return interface.make_parse(sent_len+1, c.regen(pen, counts))
def f(pen):
scorer.skip_penalty = pen
# c.score = score
parse = interface.make_parse(sent_len+1, c.regen(pen, counts))
return sent_len - parse.skipped_words()