def preload(self, sent):
self.terminalRules = DefaultDict({})
for word in sent:
wordRules = self.terminalDB[word]
for rule in wordRules:
self.terminalRules[rule.lhs][word] = rule
self.ntToWord = DefaultDict({})
for word in sent:
try:
wordLook = self.ntToWordDB[word]
for nt,prob in wordLook.items():
self.ntToWord[nt][word] = prob
except KeyError:
print >>sys.stderr, "WARNING: no word lookaheads for", word
self.posToWord = DefaultDict({})
for word in sent:
posLook = self.posToWordDB[word]
for pos,prob in posLook.items():
self.posToWord[pos][word] = prob
Grammar.preload(self, sent)
def lookaheadProbFull(self, nt, word):
presplit = nt.split("_")[0]
try:
lamb = self.lambdas[presplit]
except KeyError:
lamb = 0.5
return Grammar.lookaheadProbFull(self, nt, word, lamb=lamb)
fields = line.strip().split()
(pos, word) = fields[0:2]
lst = eval(" ".join(fields[2:]))
for num, prob in enumerate(lst):
preterm = "%s_%d" % (pos, num)
rule = Rule()
rule.setup(preterm, [word], float(prob))
if [rule.lhs] == rule.rhs and rule.prob == 1.0:
print >> sys.stderr, "Warning: X->X", rule.lhs, rule.rhs
else:
rules[rule.lhs].append(rule)
grammar = Grammar(rules)
if lookahead.endswith(".gz"):
look = GzipFile(lookahead)
else:
look = file(lookahead)
lambdas = readLambdas(look)
ntToPos = readProductionTable(look)
ntToWord = readProductionTable(look)
posToWord = readProductionTable(look)
grammar.setLookahead(lambdas, ntToPos, ntToWord, posToWord)
print >> sys.stderr, "dumping"
