def main(folder, domain, augmentation_type, num):
"""Print augmented data to stdout."""
aug_types = augmentation_type.split('+')
data = []
domain = domains.new(domain)
if folder == 'train':
fname = './geoQueryData/train/geo880_train600.tsv'
nums_data = 600
elif folder == 'dev':
fname = './geoQueryData/dev/geo880_dev100.tsv'
nums_data = 100
elif folder == 'test':
fname = './geoQueryData/test/geo880_test280.tsv'
nums_data = 280
with open(fname) as f:
for line in f:
x, y = line.strip().split('\t')
y = domain.preprocess_lf(y)
data.append((x, y))
augmenter = Augmenter(domain, data, aug_types)
aug_data = augmenter.sample(num)
path = f"./geoQueryData/{folder}/geo880_{folder}_{num + nums_data}_{augmentation_type}_recomb.tsv"
with open(path, 'w') as f:
for data in data:
f.write(data[0] + '\t' + data[1] + '\n')
for ex in aug_data:
f.write(ex[0] + '\t' + ex[1] + '\n')
def run():
configure_theano()
domain = None
if OPTIONS.domain:
domain = domains.new(OPTIONS.domain)
train_raw, dev_raw = load_raw_all(domain=domain)
random.seed(OPTIONS.model_seed)
numpy.random.seed(OPTIONS.model_seed)
#pair_stat = get_pair_stats(train_raw) # [pair co-occurances for attn2hist]
spec = init_spec(train_raw)
model = get_model(spec)
global test_domain
print('***', test_domain)
if OPTIONS.model == 'attn2hist':
model.spec.set_test_domain(test_domain)
if train_raw:
train_data = preprocess_data(model, train_raw)
#if OPTIONS.model =='attn2hist':
#set_pair_stats(model,train_data) # for the attn2hist model
#test_ibm_model(model,train_data) # for the attn2hist model
random.seed(OPTIONS.model_seed)
dev_data = None
if dev_raw:
dev_data = preprocess_data(model, dev_raw)
augmenter = get_augmenter(train_raw, domain)
model.train(train_data,
T=OPTIONS.num_epochs,
eta=OPTIONS.learning_rate,
dev_data=dev_data,
l2_reg=OPTIONS.lambda_reg,
distract_prob=OPTIONS.distract_prob,
distract_num=OPTIONS.distract_num,
concat_prob=OPTIONS.concat_prob,
concat_num=OPTIONS.concat_num,
augmenter=augmenter,
aug_frac=OPTIONS.aug_frac)
if OPTIONS.save_file:
print >> sys.stderr, 'Saving parameters...'
spec.save(OPTIONS.save_file)
#if train_raw:
# evaluate_train(model, train_data, domain=domain)
if dev_raw:
evaluate_dev(model, dev_raw, domain=domain)
write_stats()
if OPTIONS.shell:
run_shell(model)
elif OPTIONS.server:
run_server(model, hostname=OPTIONS.hostname, port=OPTIONS.port)
def run():
configure_theano()
domain = None
if OPTIONS.domain:
domain = domains.new(OPTIONS.domain)
train_raw, dev_raw = load_raw_all(domain=domain)
databases = load_databases(OPTIONS.domain_grammar, domain=domain)
random.seed(OPTIONS.model_seed)
numpy.random.seed(OPTIONS.model_seed)
spec = init_spec(train_raw, databases)
model = get_model(spec)
constructor1 = CONTROLLERS[OPTIONS.general_ontology]
constructor2 = CONTROLLERS[OPTIONS.domain_ontology]
use_general_ontology = OPTIONS.use_generalontology
use_domain_ontology = OPTIONS.use_geoontology or OPTIONS.use_atisontology or OPTIONS.use_overnightontology
general_controller = constructor1(OPTIONS.general_grammar, use_general_ontology)
domain_controller = constructor2(OPTIONS.domain_grammar, use_domain_ontology)
constructor3 = CONVERTORS[OPTIONS.domain_convertor]
domain_convertor = constructor3
if train_raw:
train_data = preprocess_data(domain_convertor, domain_controller, general_controller, model, train_raw)
random.seed(OPTIONS.model_seed)
dev_data = None
if dev_raw:
dev_data = preprocess_data(domain_convertor, domain_controller, general_controller, model, dev_raw)
augmenter = get_augmenter(train_raw, domain)
model.train(train_data, T=OPTIONS.num_epochs, eta=OPTIONS.learning_rate,
dev_data=dev_data, l2_reg=OPTIONS.lambda_reg,
distract_prob=OPTIONS.distract_prob,
distract_num=OPTIONS.distract_num,
concat_prob=OPTIONS.concat_prob, concat_num=OPTIONS.concat_num,
augmenter=augmenter, aug_frac=OPTIONS.aug_frac)
if OPTIONS.save_file:
print >> sys.stderr, 'Saving parameters...'
spec.save(OPTIONS.save_file)
evaluation_train_flag = False
if train_raw and evaluation_train_flag:
evaluate_train(model, domain_convertor, domain_controller, general_controller, train_data, domain=domain)
if dev_raw:
evaluate_dev(model, domain_convertor, domain_controller, general_controller, dev_raw, domain=domain)
write_stats()
if OPTIONS.shell:
run_shell(model)
elif OPTIONS.server:
run_server(model, hostname=OPTIONS.hostname, port=OPTIONS.port)
def evaluate_multiple_domains(name, model, dataset, domain=None):
sub_domains = {}
for ex in dataset:
sd = ex.sub_domain
if sd not in sub_domains:
sub_domains[sd] = []
sub_domains[sd].append(ex)
for sub_domain_str in sub_domains:
print sub_domain_str
sub_domain = domains.new(sub_domain_str)
evaluate(name, model, sub_domains[sub_domain_str], sub_domain)
def run():
configure_theano()
domain = None
if OPTIONS.domain:
domain = domains.new(OPTIONS.domain)
train_raw, dev_raw = load_raw_all(domain=domain)
random.seed(OPTIONS.model_seed)
numpy.random.seed(OPTIONS.model_seed)
spec = init_spec(train_raw)
model = get_model(spec)
if train_raw:
train_data = preprocess_data(model, train_raw)
random.seed(OPTIONS.model_seed)
dev_data = None
if dev_raw:
dev_data = preprocess_data(model, dev_raw)
augmenter = get_augmenter(train_raw, domain)
model.train(train_data,
T=OPTIONS.num_epochs,
eta=OPTIONS.learning_rate,
dev_data=dev_data,
l2_reg=OPTIONS.lambda_reg,
distract_prob=OPTIONS.distract_prob,
distract_num=OPTIONS.distract_num,
concat_prob=OPTIONS.concat_prob,
concat_num=OPTIONS.concat_num,
augmenter=augmenter,
aug_frac=OPTIONS.aug_frac)
if OPTIONS.save_file:
print >> sys.stderr, 'Saving parameters...'
spec.save(OPTIONS.save_file)
if train_raw:
evaluate_train(model,
train_data,
domain=domain,
sample_file='%s.train' % OPTIONS.sample_file)
if dev_raw:
evaluate_dev(model,
dev_raw,
domain=domain,
sample_file='%s.dev' % OPTIONS.sample_file)
write_stats()
if OPTIONS.shell:
run_shell(model)
elif OPTIONS.server:
run_server(model, hostname=OPTIONS.hostname, port=OPTIONS.port)
def main():
"""Print augmented data to stdout."""
if len(sys.argv) < 5:
print >> sys.stderr, 'Usage: %s [file] [domain] [aug-type] [num]' % sys.argv[
0]
sys.exit(1)
fname, domain_name, aug_type_str, num = sys.argv[1:5]
num = int(num)
aug_types = aug_type_str.split('+')
data = []
domain = domains.new(domain_name)
with open(fname) as f:
for line in f:
x, y = line.strip().split('\t')
y = domain.preprocess_lf(y)
data.append((x, y))
augmenter = Augmenter(domain, data, aug_types)
aug_data = augmenter.sample(num)
for ex in aug_data:
print '\t'.join(ex)
def main():
random.seed(0)
base_data = gen_nested()
base_train, base_test = base_data[:100], base_data[-500:]
write_data('train_base100.tsv', base_train)
write_data('test_base500.tsv', base_test)
domain = domains.new('artificial')
augmenter_entity = Augmenter(domain, base_train, ['entity'])
augmenter_nesting = Augmenter(domain, base_train, ['nesting', 'entity'])
deeper = sample_nested(depth=4, num=500)
entity_data = augmenter_entity.sample(500)
nesting_data = augmenter_nesting.sample(500)
aug_nums = (25, 50, 75, 100, 150, 200, 250, 300, 400, 500)
for n in aug_nums:
write_data('train_base%d.tsv' % (100 + n), base_data[:(100 + n)])
write_data('train_base100_entity%d.tsv' % n,
base_train + entity_data[:n])
write_data('train_base100_nesting%d.tsv' % n,
base_train + nesting_data[:n])
write_data('train_base100_deeper%d.tsv' % n, base_train + deeper[:n])
