def predict_problem(self, model_fn, test_fn, out_fn, acc_bound=0.5):
ed = self.load(model_fn)
fh = open(test_fn)
gold = []
features = []
query_string = []
for l in fh:
g, feats, query, string = l.strip().split('\t')
feats_d = dict([(int(f), 1) for f in feats.split(' ')])
gold.append(int(g))
features.append(feats_d)
query_string.append(';'.join([query, string]))
p_labels, _, p_vals = predict(gold, features, ed.model, '-b 1')
d = dict([(v, k) for k, v in ed.class_cache.iteritems()])
f = dict([(v, k) for k, v in ed.feat_cache.iteritems()])
out_fh = open(out_fn, 'w')
for i in xrange(len(p_labels)):
out_fh.write('{0}\t{1}\t{2}\t{3}\n'.format(query_string[i], d[gold[i]],
(d[int(p_labels[i])]
if p_vals[i][int(p_labels[i])] > acc_bound
else "unknown"), ';'.join([f[int(feat)]
for feat in features[i]])))
fh.close()
out_fh.close()
def predict(self,
features,
gold=None,
acc_bound=0.5,
bias=-1,
with_prob=False):
self.problem.set_bias(bias)
int_features = [self.int_feats(fvec) for fvec in features]
if gold:
gold_int_labels = [(self.class_cache[g] if type(g) == str
and g in self.class_cache else g)
for g in gold]
else:
gold_int_labels = [0 for i in xrange(len(features))]
p_labels, _, p_vals = predict(gold_int_labels, int_features,
self.model, '-b 1')
d = dict([(v, k) for k, v in self.class_cache.iteritems()])
if with_prob:
return [(d[int(p_labels[event_i])], p_vals[event_i])
for event_i in xrange(len(p_labels))]
else:
return [(d[int(p_labels[event_i])]
if p_vals[event_i][int(p_labels[event_i])] > acc_bound
else "unknown") for event_i in xrange(len(p_labels))]
def predict(self, features, gold = None):
int_features = [self.int_feats(fvec) for fvec in features]
if gold:
gold_int_labels = [self.class_cache[g] for g in gold]
else:
gold_int_labels = [0 for i in xrange(len(features))]
p_labels, _, p_vals = predict(gold_int_labels, int_features, self.model, '-b 1')
d = dict([(v, k) for k, v in self.class_cache.iteritems()])
return [d[int(label)] for label in p_labels]
def predict(self, features, gold=None, acc_bound=0.5):
int_features = [self.int_feats(fvec) for fvec in features]
if gold:
gold_int_labels = [
(self.class_cache[g] if type(g) == str
and g in self.class_cache
else g)
for g in gold]
else:
gold_int_labels = [0 for i in xrange(len(features))]
p_labels, _, p_vals = predict(gold_int_labels, int_features, self.model, '-b 1')
d = dict([(v, k) for k, v in self.class_cache.iteritems()])
return [(d[int(p_labels[event_i])]
if p_vals[event_i][int(p_labels[event_i])] > acc_bound
else "unknown")
for event_i in xrange(len(p_labels))]
def getLogTagProbsByPos(self, senFeats):
numberedSenFeats = self.getNumberedSenFeats(senFeats)
contexts = [dict([(feat, 1) for feat in feats])
for feats in numberedSenFeats]
dummyOutcomes = [1 for c in contexts]
_, __, probDistsByPos = predict(dummyOutcomes, contexts,
self.model, self.params)
"""
logTagProbsByPos = [ dict([(self.featCounter.noToFeat[i+1],
math.log(prob))
for i, prob in enumerate(probDist)])
for probDist in probDistsByPos]
"""
logTagProbsByPos = []
for probDist in probDistsByPos:
logProbsByTag = {}
for c, prob in enumerate(probDist):
tag = self.labelCounter.noToFeat[c+1]
logProbsByTag[tag] = math.log(prob)
logTagProbsByPos.append(logProbsByTag)
return logTagProbsByPos
def predict_problem(self, model_fn, test_fn, out_fn, acc_bound=0.5):
ed = self.load(model_fn)
fh = open(test_fn)
gold = []
features = []
query_string = []
for l in fh:
g, feats, query, string = l.strip().split('\t')
feats_d = dict([(int(f), 1) for f in feats.split(' ')])
gold.append(int(g))
features.append(feats_d)
query_string.append(';'.join([query, string]))
p_labels, _, p_vals = predict(gold, features, ed.model, '-b 1')
d = dict([(v, k) for k, v in ed.class_cache.iteritems()])
f = dict([(v, k) for k, v in ed.feat_cache.iteritems()])
out_fh = open(out_fn, 'w')
for i in xrange(len(p_labels)):
out_fh.write('{0}\t{1}\t{2}\t{3}\n'.format(
query_string[i], d[gold[i]],
(d[int(p_labels[i])]
if p_vals[i][int(p_labels[i])] > acc_bound else "unknown"),
';'.join([f[int(feat)] for feat in features[i]])))
fh.close()
out_fh.close()
