tm = models.TM(opts.tm, sys.maxint)
lm = models.LM(opts.lm)
french_sents = [tuple(line.strip().split()) for line in open(opts.input).readlines()[:opts.num_sents]]
outputs = glob.glob("./old/out*")
results = collections.defaultdict(list)
count = 0
for output in outputs:
count = count + 1
out = [tuple(line.strip().split()) for line in open(output).readlines()]
if (len(french_sents) != len(out)): continue
sent_num = 0
for e in out:
prob = gradeOneSent.gradeOneSentence(sent_num, tuple(e), french_sents[sent_num], lm, tm);
results[sent_num].append((prob, e))
sent_num = sent_num + 1
def printTuple(t):
s = ""
for i in range(len(t)):
s = s + t[i] + " "
return s
for sent_num in range(55):
hypos = sorted(results[sent_num], key=lambda tup: tup[0], reverse=True)
print printTuple(hypos[0][1])
# on the last stack
winner = max(stacks[-1].itervalues(), key=lambda h: h.logprob)
def extract_english_recursive(h):
return '' if h.predecessor is None else '%s%s ' % (extract_english_recursive(h.predecessor), h.phrase.english)
def extract_tm_logprob(h):
return 0.0 if h.predecessor is None else h.phrase.logprob + extract_tm_logprob(h.predecessor)
sent_num = sent_num +1
# coding for Marginalization
probMax = float("-inf")
#for h0 in heapq.nlargest(1000, stacks[-1].itervalues(), key=lambda h: h.logprob): # TODO: only consider 1000 top hypothesis in the last stack
for h0 in stacks[-1].itervalues(): # consider all the hypothesis in the last stack
e = tuple(extract_english_recursive(h0).strip().split())
prob = gradeOneSent.gradeOneSentence(sent_num, e, f, lm, tm)
if prob > probMax:
eMax = e
probMax = prob
e = eMax
# do all the reordering
for i in range(len(e)-1):
#k = min(i+20,len(e)-1) #TODO:
k = len(e)-1
for j in range(i+1, k):
etmp = list(e)
etmp[i] = e[j]
etmp[j] = e[i]
etmp = tuple(etmp)