def main(transductive: bool = False):
try:
from classifier import pretrain
except ImportError:
part2xy = load_dataset_fast('FILIMDB', parts=SCORED_PARTS)
train_ids, train_texts, train_labels = part2xy['train']
print('\nTraining classifier on %d examples from train set ...' %
len(train_texts))
st = time()
params = train(train_texts, train_labels)
print('Classifier trained in %.2fs' % (time() - st))
else:
part2xy = load_dataset_fast('FILIMDB',
parts=SCORED_PARTS + ('train_unlabeled', ))
train_ids, train_texts, train_labels = part2xy['train']
_, train_unlabeled_texts, _ = part2xy['train_unlabeled']
st = time()
if transductive:
all_texts = list(text for _, text, _ in part2xy.values())
else:
all_texts = [train_texts, train_unlabeled_texts]
total_texts = sum(len(text) for text in all_texts)
print('\nPretraining classifier on %d examples' % total_texts)
params = pretrain(all_texts)
print('Classifier pretrained in %.2fs' % (time() - st))
print('\nTraining classifier on %d examples from train set ...' %
len(train_texts))
st = time()
params = train(train_texts, train_labels, params)
print('Classifier trained in %.2fs' % (time() - st))
del part2xy["train_unlabeled"]
allpreds = []
for part, (ids, x, y) in part2xy.items():
print('\nClassifying %s set with %d examples ...' % (part, len(x)))
st = time()
preds = classify(x, params)
print('%s set classified in %.2fs' % (part, time() - st))
allpreds.extend(zip(ids, preds))
if y is None:
print('no labels for %s set' % part)
else:
score(preds, y)
save_preds(allpreds, preds_fname=PREDS_FNAME)
print('\nChecking saved predictions ...')
score_preds(preds_fname=PREDS_FNAME, data_dir='FILIMDB')
def main():
part2xy = load_dataset_fast('FILIMDB')
train_ids, train_texts, train_labels = part2xy['train']
print('\nTraining classifier on %d examples from train set ...' %
len(train_texts))
st = time()
params = train(train_texts, train_labels)
print('Classifier trained in %.2fs' % (time() - st))
allpreds = []
for part, (ids, x, y) in part2xy.items():
print('\nClassifying %s set with %d examples ...' % (part, len(x)))
st = time()
preds = classify(x, params)
print('%s set classified in %.2fs' % (part, time() - st))
allpreds.extend(zip(ids, preds))
if y is None:
print('no labels for %s set' % part)
else:
score(preds, y)
save_preds(allpreds, preds_fname=PREDS_FNAME)
print('\nChecking saved predictions ...')
score_preds(preds_fname=PREDS_FNAME)
def main():
try:
from classifier import pretrain
except ImportError:
part2xy = load_dataset_fast('FILIMDB')
train_ids, train_texts, train_labels = part2xy['train']
print('\nTraining classifier on %d examples from train set ...' %
len(train_texts))
st = time()
params = train(train_texts, train_labels)
print('Classifier trained in %.2fs' % (time() - st))
else:
part2xy = load_dataset_fast('FILIMDB',
parts=('train', 'dev', 'test',
'train_unlabeled'))
train_ids, train_texts, train_labels = part2xy['train']
_, train_unlabeled_texts, _ = part2xy['train_unlabeled']
all_texts = train_texts + train_unlabeled_texts
print('\nPretraining classifier on %d examples' % len(all_texts))
st = time()
params = pretrain(all_texts)
print('Classifier pretrained in %.2fs' % (time() - st))
print('\nTraining classifier on %d examples from train set ...' %
len(train_texts))
st = time()
params = train(train_texts, train_labels, params)
print('Classifier trained in %.2fs' % (time() - st))
del part2xy["train_unlabeled"]
allpreds = []
for part, (ids, x, y) in part2xy.items():
print('\nClassifying %s set with %d examples ...' % (part, len(x)))
st = time()
preds = classify(x, params)
print('%s set classified in %.2fs' % (part, time() - st))
allpreds.extend(zip(ids, preds))
if y is None:
print('no labels for %s set' % part)
else:
score(preds, y)
save_preds(allpreds, preds_fname=PREDS_FNAME)
print('\nChecking saved predictions ...')
score_preds(preds_fname=PREDS_FNAME, data_dir='FILIMDB')
def pretrain(ds_name, module, part2xy, transductive):
train_ids, train_texts, train_labels = part2xy['train']
_, train_unlabeled_texts, _ = score.load_dataset_fast(ds_name, parts=('train_unlabeled',))['train_unlabeled']
if transductive:
all_texts = [text for _, text, _ in part2xy.values()] + [train_unlabeled_texts]
else:
all_texts = [train_texts, train_unlabeled_texts]
total_texts = sum(len(text) for text in all_texts)
# print('\nPretraining classifier on %d examples from %s; transductive=%s' % (total_texts, pretrain_parts,transductive))
print('\nPretraining classifier on %d examples' % total_texts)
st = time()
params = module.pretrain(all_texts)
print('Classifier pretrained in %.2fs' % (time() - st))
return params
for word in doc_set:
k = doc.count(word)
if word not in vocab:
p_pos = alpha / (alpha * (vocab_size + 1) + params.all_pos)
p_neg = alpha / (alpha * (vocab_size + 1) + params.all_neg)
else:
p_pos = params.pos_prob[word]
p_neg = params.neg_prob[word]
k_log_fact = math.log(math.factorial(k))
pos_bayes += k * math.log(p_pos) - k_log_fact
neg_bayes += k * math.log(p_neg) - k_log_fact
pos_bayes += math.log(params.prob_of_pos_class)
neg_bayes += math.log(params.prob_of_neg_class)
if pos_bayes > neg_bayes:
labels.append('pos')
else:
labels.append('neg')
return labels
t0 = time.clock()
text = score.load_dataset_fast()
train_ids, train_texts, train_labels = text['train']
test_ids, test_texts, test_labels = text['dev']
w = train(train_texts, train_labels)
print(w)
print("Training time: %f" % time.clock())
def main(train_timeout=5 * 60, eval_timeout=5 * 60):
results = {}
try:
import classifier
importlib.reload(classifier)
except Exception as e:
print(e)
results["exception"] = str(e)
if sys.modules.get("classifier"):
del sys.modules['classifier']
return results
part2xy = load_dataset_fast('FILIMDB_hidden', SCORED_PARTS)
train_ids, train_texts, train_labels = part2xy['train']
print('\nTraining classifier on %d examples from train set ...' % len(train_texts))
st = time()
try:
with time_limit(train_timeout):
params = classifier.train(train_texts, train_labels)
except (TimeoutException, ValueError, Exception) as e:
del sys.modules['classifier']
print(e)
if isinstance(e, TimeoutException):
results["train_time"] = train_timeout
results["exception"] = str(e)
return results
train_time = time() - st
results["train_time"] = train_time
print('Classifier trained in %.2fs' % train_time)
allpreds = []
for part, (ids, x, y) in part2xy.items():
print('\nClassifying %s set with %d examples ...' % (part, len(x)))
st = time()
try:
with time_limit(eval_timeout):
preds = classifier.classify(x, params)
except (TimeoutException, ValueError) as e:
del sys.modules['classifier']
if isinstance(e, TimeoutException):
print("Timeout on evaluating %s set!" % part)
results["eval_on_%s_set_time" % part] = eval_timeout
else:
print(e)
results["exception"] = str(e)
return results
eval_time = time() - st
results["eval_on_%s_set_time" % part] = eval_time
print('%s set classified in %.2fs' % (part, eval_time))
allpreds.extend(zip(ids, preds))
if y is None:
print('no labels for %s set' % part)
else:
acc = score(preds, y)
results["eval_on_%s_set_acc" % part] = acc
del sys.modules['classifier']
return results
def load_ds(ds_name: str):
return score.load_dataset_fast(ds_name, parts=score.SCORED_PARTS)