def test_get_most_common_tag():
expected = 0.63
weights = most_common.get_most_common_weights(TRAIN_FILE)
confusion = tagger_base.evalTagger(tagger_base.makeClassifierTagger(weights),'mcc')
actual = scorer.accuracy(confusion)
ok_(expected < actual, msg="NOT_IN_RANGE Expected:%f, Actual:%f" %(expected, actual))
def trainAvgStructPerceptron(N_its,inst_generator,featfunc,tagset):
"""
:param N_its: number of iterations
:param inst_generator: A generator of (words,tags) tuples
:param tagger: A function from (words, weights) to tags
:param features: A function from (words, tags) to a dict of features and weights
"""
tr_acc = [None]*N_its
dv_acc = [None]*N_its
T = 0
weights = defaultdict(float)
wsum = defaultdict(float)
avg_weights = defaultdict(float)
for i in xrange(N_its):
# your code here
weights, wsum, tr_acc_i, num_instances = oneItAvgStructPerceptron(inst_generator, featfunc, weights, wsum, tagset, T)
# note that I call evalTagger to produce the dev set results
T += num_instances
for w in wsum:
avg_weights[w] = weights[w] - wsum[w] / float(T)
confusion = evalTagger(lambda words,tags : viterbiTagger(words,featfunc,avg_weights,tags)[0],'sp.txt')
dv_acc[i] = scorer.accuracy(confusion)
tr_acc[i] = tr_acc_i#1. - tr_err/float(sum([len(s) for s,t in inst_generator]))
print i,'dev:',dv_acc[i],'train:',tr_acc[i]
return avg_weights, tr_acc, dv_acc
def trainAvgPerceptron(N_its,inst_generator,featfunc,tagset):
"""
:param N_its: number of iterations
:param inst_generator: generate words,tags pairs
:param featfunc: feature function
:param tagset: set of all possible tags
:returns average weights, training accuracy, dev accuracy
"""
tr_acc = [None]*N_its
dv_acc = [None]*N_its
T = 0
weights = defaultdict(float)
wsum = defaultdict(float)
avg_weights = defaultdict(float)
for i in xrange(N_its):
# your code here
weights, wsum, tr_acc_i, num_instances = oneItAvgPerceptron(inst_generator, featfunc, weights, wsum, tagset, T)
T += num_instances
for w in wsum:
avg_weights[w] = weights[w] - wsum[w] / float(T)
confusion = evalTagger(lambda words, alltags: classifierTagger(words,featfunc,avg_weights,tagset),'perc')
dv_acc[i] = scorer.accuracy(confusion)
tr_acc[i] = tr_acc_i
print i,'dev:',dv_acc[i],'train:',tr_acc[i]
return avg_weights, tr_acc, dv_acc
def trainAvgStructPerceptron(N_its, inst_generator, featfunc, tagset):
"""
:param N_its: number of iterations
:param inst_generator: A generator of (words,tags) tuples
:param tagger: A function from (words, weights) to tags
:param features: A function from (words, tags) to a dict of features and weights
"""
tr_acc = [None] * N_its
dv_acc = [None] * N_its
T = 0
weights = defaultdict(float)
wsum = defaultdict(float)
avg_weights = defaultdict(float)
for i in xrange(N_its):
# your code here
weights, wsum, tr_acc_i, num_instances = oneItAvgStructPerceptron(
inst_generator, featfunc, weights, wsum, tagset, T)
# note that I call evalTagger to produce the dev set results
T += num_instances
for w in wsum:
avg_weights[w] = weights[w] - wsum[w] / float(T)
confusion = evalTagger(
lambda words, tags: viterbiTagger(words, featfunc, avg_weights,
tags)[0], 'sp.txt')
dv_acc[i] = scorer.accuracy(confusion)
tr_acc[
i] = tr_acc_i #1. - tr_err/float(sum([len(s) for s,t in inst_generator]))
print i, 'dev:', dv_acc[i], 'train:', tr_acc[i]
return avg_weights, tr_acc, dv_acc
def test_hmm_weights_accuracy():
confusion = tagger_base.evalTagger(
lambda words, alltags: viterbi.viterbiTagger(
words, viterbi.hmm_feats, hmm_weights, alltags)[0], 'hmm')
actual = scorer.accuracy(confusion)
expected = 0.74
ok_(expected < actual,
msg="NOT_IN_RANGE Expected:%f, Actual:%f" % (expected, actual))
def test_get_most_common_tag():
expected = 0.63
weights = most_common.get_most_common_weights(TRAIN_FILE)
confusion = tagger_base.evalTagger(
tagger_base.makeClassifierTagger(weights), 'mcc')
actual = scorer.accuracy(confusion)
ok_(expected < actual,
msg="NOT_IN_RANGE Expected:%f, Actual:%f" % (expected, actual))
def test_classifier_tagger():
expected = 0.136844287788
noun_weights = most_common.get_noun_weights()
noun_tagger = tagger_base.makeClassifierTagger(noun_weights)
confusion = tagger_base.evalTagger(noun_tagger,'nouns')
actual = scorer.accuracy(confusion)
assert_almost_equals(expected, actual,places=3, msg="UNEQUAL Expected:%s, Actual:%s" %(expected, actual))
def test_str_perceptron_small():
w, tr_acc, dv_acc = str_perceptron.trainAvgStructPerceptron(
5, tr_all[:50], features.wordTransFeatures, alltags)
confusion = tagger_base.evalTagger(
lambda words, alltags: viterbi.viterbiTagger(
words, features.wordTransFeatures, w, alltags)[0],
'str_classifier_small')
expected_acc = 0.506
actual_acc = scorer.accuracy(confusion)
ok_(expected_acc < actual_acc,
msg="NOT_IN_RANGE Expected:%f, Actual:%f" % (expected_acc, actual_acc))
def test_classifier_tagger():
expected = 0.136844287788
noun_weights = most_common.get_noun_weights()
noun_tagger = tagger_base.makeClassifierTagger(noun_weights)
confusion = tagger_base.evalTagger(noun_tagger, 'nouns')
actual = scorer.accuracy(confusion)
assert_almost_equals(expected,
actual,
places=3,
msg="UNEQUAL Expected:%s, Actual:%s" %
(expected, actual))
def test_basic_classifer():
test_weights = defaultdict(float)
test_tags = ['N','V','V','N']
for i in range(len(sent)):
for feat in features.wordFeatures(sent,test_tags[i],'X',i):
test_weights[feat] = 1
for feat in features.wordFeatures(sent,'X','X',i):
test_weights[feat] = 1
expected = test_tags
actual = tagger_base.classifierTagger(sent,features.wordFeatures,test_weights,alltags)
eq_ (expected, actual, msg="UNEQUAL Expected:%s, Actual:%s" %(expected, actual) )
expected_acc = 0.139539705577
confusion = tagger_base.evalTagger(lambda words,alltags : tagger_base.classifierTagger(words,features.wordFeatures,test_weights,alltags),'test')
actual_acc =scorer.accuracy(confusion)
assert_almost_equals(expected_acc ,actual_acc,places = 3)
def test_basic_classifer():
test_weights = defaultdict(float)
test_tags = ['N', 'V', 'V', 'N']
for i in range(len(sent)):
for feat in features.wordFeatures(sent, test_tags[i], 'X', i):
test_weights[feat] = 1
for feat in features.wordFeatures(sent, 'X', 'X', i):
test_weights[feat] = 1
expected = test_tags
actual = tagger_base.classifierTagger(sent, features.wordFeatures,
test_weights, alltags)
eq_(expected,
actual,
msg="UNEQUAL Expected:%s, Actual:%s" % (expected, actual))
expected_acc = 0.139539705577
confusion = tagger_base.evalTagger(
lambda words, alltags: tagger_base.classifierTagger(
words, features.wordFeatures, test_weights, alltags), 'test')
actual_acc = scorer.accuracy(confusion)
assert_almost_equals(expected_acc, actual_acc, places=3)
def test_hmm_weights_accuracy():
confusion = tagger_base.evalTagger(lambda words, alltags : viterbi.viterbiTagger(words,viterbi.hmm_feats,hmm_weights,alltags)[0],'hmm')
actual = scorer.accuracy(confusion)
expected = 0.74
ok_ (expected < actual, msg="NOT_IN_RANGE Expected:%f, Actual:%f" %(expected, actual))
def test_str_perceptron_small():
w,tr_acc,dv_acc = str_perceptron.trainAvgStructPerceptron(5,tr_all[:50],features.wordTransFeatures,alltags)
confusion = tagger_base.evalTagger(lambda words,alltags : viterbi.viterbiTagger(words,features.wordTransFeatures,w,alltags)[0],'str_classifier_small')
expected_acc = 0.506
actual_acc = scorer.accuracy(confusion)
ok_ (expected_acc < actual_acc, msg="NOT_IN_RANGE Expected:%f, Actual:%f" %(expected_acc, actual_acc))
