def main():
parser = argparse.ArgumentParser(description='')
parser.add_argument('-i', '--input', help='Input file', required=True)
parser.add_argument('-t', '--train', help='Additional training only data')
parser.add_argument('-f',
'--folds',
help='Number of folds (default: 50)',
type=int,
default=50)
parser.add_argument('-s',
'--seed',
help='Randomization seed (default: 0)',
type=int,
default=0)
parser.add_argument('-v',
'--verbose',
help='Verbose output (default: no)',
action='store_true')
parser.add_argument('-r',
'--reference_label',
help='Name of label to be quantified',
type=str,
default='positive')
parser.add_argument('-k',
'--k',
help='Number of features to keep',
type=int,
default=1000)
parser.add_argument('-c',
'--c',
help='C factor for svm (default: 1.0)',
type=float,
default=1.0)
parser.add_argument(
'-l',
'--learner',
help='Base learner to use (defalut:svm, options: svm, lr)',
type=str,
default='svm')
args = parser.parse_args()
if args.verbose:
logging.basicConfig(level=logging.DEBUG)
data = read_semeval_quantification_classification(args.input,
encoding='windows-1252',
verbose=args.verbose)
if args.train:
train = read_semeval_quantification_classification(
args.train, encoding='windows-1252', verbose=args.verbose)
else:
train = None
print(
'"Test set","Category","True prevalence","Method","Predicted prevalence"'
)
for test_key in data:
tests = list()
X = list()
y = list()
for key in data:
if key == test_key:
tests.append([key, data[key][0], data[key][1]])
else:
X.extend(data[key][0])
y.extend(data[key][1])
if train is not None:
for key in train:
X.extend(train[key][0])
y.extend(train[key][1])
logging.debug('Training set size %i' % len(y))
logging.debug('Number of test sets %i' % len(tests))
for test in tests:
logging.debug('Test %s size %i' % (test[0], len(test[1])))
if args.learner == 'svm':
learner = svm.SVC(kernel='linear', probability=True, C=args.c)
else:
learner = linear_model.LogisticRegression(C=args.c)
quantifier = Quantifier(learner,
reference_label=args.reference_label,
n_folds=args.folds,
seed=args.seed)
pipeline = Pipeline([
('vect', CountVectorizer()),
('tfidf', TfidfTransformer()),
('chi2', SelectKBest(chi2, k=args.k)),
('clf', quantifier),
])
quantifier = pipeline.fit(X, y)
true_prevalences = dict()
results = dict()
for testname, texts, labels in tests:
true_prevalences[testname] = labels.count(
args.reference_label) / len(labels)
results[testname] = quantifier.predict(texts)
for testname in results:
for result in results[testname]:
print('"%s","%s",%0.3f,"%s",%0.3f' %
(testname, result[1], true_prevalences[testname],
result[0], result[2]))
def main():
parser = argparse.ArgumentParser(description='')
parser.add_argument('-i', '--input', help='Input file', required=True)
parser.add_argument('-r', '--train', help='Additional training only data')
parser.add_argument('-f', '--folds', help='Number of folds (default: leave one out)', type=int)
parser.add_argument('-s', '--seed', help='Randomization seed (default: 0)', type=int, default=0)
parser.add_argument('-v', '--verbose', help='Verbose output (default: no)', action='store_true')
parser.add_argument('-l', '--reference_label', help='Name of label to be quantified', type=str, default='positive')
parser.add_argument('-t', '--test', help='Test file', required=True)
parser.add_argument('-c', '--c', help='C value for SVM', type=float, default=1.0)
parser.add_argument('-k', '--k', help='Number of features to keep', type=int, default=1000)
parser.add_argument('-o', '--output', help='Output filename prefix', required=True)
args = parser.parse_args()
if args.verbose:
logging.basicConfig(level=logging.DEBUG)
data = read_semeval_quantification_classification(args.input, encoding='windows-1252', verbose=args.verbose)
if args.train:
train = read_semeval_quantification_classification(args.train, encoding='windows-1252', verbose=args.verbose)
else:
train = None
test = read_semeval_quantification_classification(args.test, encoding='windows-1252', verbose=args.verbose,
delimiter='\t')
X = list()
y = list()
for key in data:
X.extend(data[key][0])
y.extend(data[key][1])
if train is not None:
for key in train:
X.extend(train[key][0])
y.extend(train[key][1])
logging.debug('Training set size %i' % len(y))
logging.debug('Number of test sets %i' % len(test))
learner = SVC(C=args.c, kernel='linear', probability=True)
if args.folds:
logging.debug('Folds %i' % args.folds)
quantifier = Quantifier(learner, reference_label=args.reference_label,
n_folds=args.folds,
seed=args.seed)
else:
logging.debug('Leave one out %i' % len(y))
quantifier = Quantifier(learner, reference_label=args.reference_label,
n_folds=len(y),
seed=args.seed)
analyzer = get_rich_analyzer(word_ngrams=[2, 3], char_ngrams=[4])
pipeline = Pipeline([
('vect', CountVectorizer(analyzer=analyzer)),
('tfidf', TfidfTransformer()),
('chi2', SelectKBest(chi2, k=args.k)),
('clf', quantifier),
])
quantifier = pipeline.fit(X, y)
true_prevalences = dict()
results = dict()
for key in test:
texts, labels = test[key]
true_prevalences[key] = labels.count(args.reference_label) / len(labels)
results[key] = quantifier.predict(texts)
with open('%sc%f-k%i-cc-D.output' % (args.output, args.c, args.k), 'w', encoding='utf8') as ccfile, \
open('%sc%f-k%i-acc-D.output' % (args.output, args.c, args.k), 'w', encoding='utf8') as accfile, \
open('%sc%f-k%i-pcc-D.output' % (args.output, args.c, args.k), 'w', encoding='utf8') as pccfile, \
open('%sc%f-k%i-pacc-D.output' % (args.output, args.c, args.k), 'w', encoding='utf8') as paccfile:
topics = list(results)
topics.sort()
for topic in topics:
for result in results[topic]:
if result[0] == 'CC':
print('%s\t%0.3f\t%0.3f' % (topic, result[2], 1 - result[2]), file=ccfile)
elif result[0] == 'ACC':
print('%s\t%0.3f\t%0.3f' % (topic, result[2], 1 - result[2]), file=accfile)
elif result[0] == 'PCC':
print('%s\t%0.3f\t%0.3f' % (topic, result[2], 1 - result[2]), file=pccfile)
elif result[0] == 'PACC':
print('%s\t%0.3f\t%0.3f' % (topic, result[2], 1 - result[2]), file=paccfile)
def main():
parser = argparse.ArgumentParser(description='')
parser.add_argument('-i', '--input', help='Input file', required=True)
parser.add_argument('-r', '--train', help='Additional training only data')
parser.add_argument('-f',
'--folds',
help='Number of folds (default: leave one out)',
type=int)
parser.add_argument('-s',
'--seed',
help='Randomization seed (default: 0)',
type=int,
default=0)
parser.add_argument('-v',
'--verbose',
help='Verbose output (default: no)',
action='store_true')
parser.add_argument('-l',
'--reference_label',
help='Name of label to be quantified',
type=str,
default='positive')
parser.add_argument('-t', '--test', help='Test file', required=True)
parser.add_argument('-c',
'--c',
help='C value for SVM',
type=float,
default=1.0)
parser.add_argument('-k',
'--k',
help='Number of features to keep',
type=int,
default=1000)
parser.add_argument('-o',
'--output',
help='Output filename prefix',
required=True)
args = parser.parse_args()
if args.verbose:
logging.basicConfig(level=logging.DEBUG)
data = read_semeval_quantification_classification(args.input,
encoding='windows-1252',
verbose=args.verbose)
if args.train:
train = read_semeval_quantification_classification(
args.train, encoding='windows-1252', verbose=args.verbose)
else:
train = None
test = read_semeval_quantification_classification(args.test,
encoding='windows-1252',
verbose=args.verbose,
delimiter='\t')
X = list()
y = list()
for key in data:
X.extend(data[key][0])
y.extend(data[key][1])
if train is not None:
for key in train:
X.extend(train[key][0])
y.extend(train[key][1])
logging.debug('Training set size %i' % len(y))
logging.debug('Number of test sets %i' % len(test))
learner = SVC(C=args.c, kernel='linear', probability=True)
if args.folds:
logging.debug('Folds %i' % args.folds)
quantifier = Quantifier(learner,
reference_label=args.reference_label,
n_folds=args.folds,
seed=args.seed)
else:
logging.debug('Leave one out %i' % len(y))
quantifier = Quantifier(learner,
reference_label=args.reference_label,
n_folds=len(y),
seed=args.seed)
analyzer = get_rich_analyzer(word_ngrams=[2, 3], char_ngrams=[4])
pipeline = Pipeline([
('vect', CountVectorizer(analyzer=analyzer)),
('tfidf', TfidfTransformer()),
('chi2', SelectKBest(chi2, k=args.k)),
('clf', quantifier),
])
quantifier = pipeline.fit(X, y)
true_prevalences = dict()
results = dict()
for key in test:
texts, labels = test[key]
true_prevalences[key] = labels.count(
args.reference_label) / len(labels)
results[key] = quantifier.predict(texts)
with open('%sc%f-k%i-cc-D.output' % (args.output, args.c, args.k), 'w', encoding='utf8') as ccfile, \
open('%sc%f-k%i-acc-D.output' % (args.output, args.c, args.k), 'w', encoding='utf8') as accfile, \
open('%sc%f-k%i-pcc-D.output' % (args.output, args.c, args.k), 'w', encoding='utf8') as pccfile, \
open('%sc%f-k%i-pacc-D.output' % (args.output, args.c, args.k), 'w', encoding='utf8') as paccfile:
topics = list(results)
topics.sort()
for topic in topics:
for result in results[topic]:
if result[0] == 'CC':
print('%s\t%0.3f\t%0.3f' %
(topic, result[2], 1 - result[2]),
file=ccfile)
elif result[0] == 'ACC':
print('%s\t%0.3f\t%0.3f' %
(topic, result[2], 1 - result[2]),
file=accfile)
elif result[0] == 'PCC':
print('%s\t%0.3f\t%0.3f' %
(topic, result[2], 1 - result[2]),
file=pccfile)
elif result[0] == 'PACC':
print('%s\t%0.3f\t%0.3f' %
(topic, result[2], 1 - result[2]),
file=paccfile)
def main():
parser = argparse.ArgumentParser(description='')
parser.add_argument('-i', '--input', help='Input file', required=True)
parser.add_argument('-t', '--train', help='Additional training only data')
parser.add_argument('-f', '--folds', help='Number of folds (default: 50)', type=int, default=50)
parser.add_argument('-s', '--seed', help='Randomization seed (default: 0)', type=int, default=0)
parser.add_argument('-v', '--verbose', help='Verbose output (default: no)', action='store_true')
parser.add_argument('-r', '--reference_label', help='Name of label to be quantified', type=str, default='positive')
parser.add_argument('-k', '--k', help='Number of features to keep', type=int, default=1000)
parser.add_argument('-c', '--c', help='C factor for svm (default: 1.0)', type=float, default=1.0)
parser.add_argument('-l', '--learner', help='Base learner to use (defalut:svm, options: svm, lr)', type=str,
default='svm')
args = parser.parse_args()
if args.verbose:
logging.basicConfig(level=logging.DEBUG)
data = read_semeval_quantification_classification(args.input, encoding='windows-1252', verbose=args.verbose)
if args.train:
train = read_semeval_quantification_classification(args.train, encoding='windows-1252', verbose=args.verbose)
else:
train = None
print('"Test set","Category","True prevalence","Method","Predicted prevalence"')
for test_key in data:
tests = list()
X = list()
y = list()
for key in data:
if key == test_key:
tests.append([key, data[key][0], data[key][1]])
else:
X.extend(data[key][0])
y.extend(data[key][1])
if train is not None:
for key in train:
X.extend(train[key][0])
y.extend(train[key][1])
logging.debug('Training set size %i' % len(y))
logging.debug('Number of test sets %i' % len(tests))
for test in tests:
logging.debug('Test %s size %i' % (test[0], len(test[1])))
if args.learner == 'svm':
learner = svm.SVC(kernel='linear', probability=True, C=args.c)
else:
learner = linear_model.LogisticRegression(C=args.c)
quantifier = Quantifier(learner, reference_label=args.reference_label,
n_folds=args.folds,
seed=args.seed)
pipeline = Pipeline([
('vect', CountVectorizer()),
('tfidf', TfidfTransformer()),
('chi2', SelectKBest(chi2, k=args.k)),
('clf', quantifier),
])
quantifier = pipeline.fit(X, y)
true_prevalences = dict()
results = dict()
for testname, texts, labels in tests:
true_prevalences[testname] = labels.count(args.reference_label) / len(labels)
results[testname] = quantifier.predict(texts)
for testname in results:
for result in results[testname]:
print('"%s","%s",%0.3f,"%s",%0.3f' % (
testname, result[1], true_prevalences[testname], result[0], result[2]))
