def get_dataset_from_path(path_dataset, vocab=None, shrink=1,
char_based=False):
train = read_first_col_is_label_format(os.path.join(path_dataset, 'train'),
char_based=char_based)
test = read_first_col_is_label_format(os.path.join(path_dataset, 'test'),
char_based=char_based)
train = transform_to_array(train, vocab)
test = transform_to_array(test, vocab)
return train, test, vocab
def get_dataset_from_path(path_dataset,
vocab=None,
shrink=1,
char_based=False):
train = read_other_dataset(os.path.join(path_dataset, 'train'),
shrink=shrink,
char_based=char_based)
test = read_other_dataset(os.path.join(path_dataset, 'test'),
shrink=shrink,
char_based=char_based)
train = transform_to_array(train, vocab)
test = transform_to_array(test, vocab)
return train, test, vocab
def predict(model, sentence):
model, vocab, setup = model
sentence = sentence.strip()
text = nlp_utils.normalize_text(sentence)
words = nlp_utils.split_text(text, char_based=setup['char_based'])
xs = nlp_utils.transform_to_array([words], vocab, with_label=False)
xs = nlp_utils.convert_seq(xs, device=-1, with_label=False) # todo use GPU
with chainer.using_config('train', False), chainer.no_backprop_mode():
prob = model.predict(xs, softmax=True)[0]
answer = int(model.xp.argmax(prob))
score = float(prob[answer])
return answer, score
def get_vectors(model, sentences):
model, vocab, setup = model
vectors = []
for sentence in sentences:
sentence = sentence.strip()
text = nlp_utils.normalize_text(sentence)
words = nlp_utils.split_text(text, char_based=setup['char_based'])
xs = nlp_utils.transform_to_array([words], vocab, with_label=False)
xs = nlp_utils.convert_seq(xs, device=-1,
with_label=False) # todo use GPU
with chainer.using_config('train', False), chainer.no_backprop_mode():
vector = model.encoder(xs)
vectors.append(vector.data[0])
vectors = numpy.asarray(vectors)
return vectors
def get_result(self,
embeddings,
path_dataset,
path_output='/tmp/text_classification/'):
self.out = path_output
self.unit = embeddings.matrix.shape[1]
if not os.path.isdir(path_output):
os.makedirs(path_output)
# TODO: move this to protonn ds management
self.path_dataset = path_dataset
# if self.path_dataset == 'dbpedia':
# train, test, vocab = text_datasets.get_dbpedia(
# char_based=self.char_based,
# vocab=embeddings.vocabulary.dic_words_ids,
# shrink=self.shrink)
# elif self.path_dataset.startswith('imdb.'):
# train, test, vocab = text_datasets.get_imdb(
# fine_grained=self.path_dataset.endswith('.fine'),
# char_based=self.char_based,
# vocab=embeddings.vocabulary.dic_words_ids,
# shrink=self.shrink)
# elif self.path_dataset in ['TREC', 'stsa.binary', 'stsa.fine',
# 'custrev', 'mpqa', 'rt-polarity', 'subj']:
# train, test, vocab = text_datasets.get_other_text_dataset(
# self.path_dataset,
# char_based=self.char_based,
# vocab=embeddings.vocabulary.dic_words_ids,
# shrink=self.shrink)
# else: # finallly, if file is not downloadable, load from local path
print(path_dataset)
path_adapter = os.path.join(path_dataset, "adapter.py")
if os.path.isfile(path_adapter):
spec = importlib.util.spec_from_file_location(
"ds_adapter", path_adapter)
module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(module)
adapter = module.Adapter()
train, test, _ = adapter.read()
vocab = embeddings.vocabulary.dic_words_ids
train = nlp_utils.transform_to_array(train, vocab)
test = nlp_utils.transform_to_array(test, vocab)
# exit(0)
else:
train, test, vocab = text_datasets.get_dataset_from_path(
path_dataset,
vocab=embeddings.vocabulary.dic_words_ids,
char_based=self.char_based,
shrink=self.shrink)
print('# cnt train samples: {}'.format(len(train)))
print('# cnt test samples: {}'.format(len(test)))
print('# size vocab: {}'.format(len(vocab)))
n_class = len(set([int(d[1]) for d in train]))
print('# cnt classes: {}'.format(n_class))
# print(train[0])
# exit(0)
train_iter = chainer.iterators.SerialIterator(train, self.batchsize)
test_iter = chainer.iterators.SerialIterator(test,
self.batchsize,
repeat=False,
shuffle=False)
# Setup a model
if self.model == 'rnn':
Encoder = nets.RNNEncoder
elif self.model == 'cnn':
Encoder = nets.CNNEncoder
elif self.model == 'bow':
Encoder = nets.BOWMLPEncoder
encoder = Encoder(n_layers=self.layer,
n_vocab=len(vocab),
n_units=self.unit,
dropout=self.dropout,
wv=embeddings.matrix)
model = nets.TextClassifier(encoder, n_class)
if self.gpu >= 0:
# Make a specified GPU current
chainer.backends.cuda.get_device_from_id(self.gpu).use()
model.to_gpu() # Copy the model to the GPU
# Setup an optimizer
optimizer = chainer.optimizers.Adam()
optimizer.setup(model)
optimizer.add_hook(chainer.optimizer.WeightDecay(1e-4))
# Set up a trainer
updater = training.StandardUpdater(train_iter,
optimizer,
converter=nlp_utils.convert_seq,
device=self.gpu)
trainer = training.Trainer(updater, (self.epoch, 'epoch'),
out=self.out)
# Evaluate the model with the test dataset for each epoch
trainer.extend(
extensions.Evaluator(test_iter,
model,
converter=nlp_utils.convert_seq,
device=self.gpu))
# Take a best snapshot
record_trigger = training.triggers.MaxValueTrigger(
'validation/main/accuracy', (1, 'epoch'))
trainer.extend(extensions.snapshot_object(model, 'best_model.npz'),
trigger=record_trigger)
# Write a log of evaluation statistics for each epoch
trainer.extend(extensions.LogReport())
trainer.extend(
extensions.PrintReport([
'epoch', 'main/loss', 'validation/main/loss', 'main/accuracy',
'validation/main/accuracy', 'elapsed_time'
]))
# Print a progress bar to stdout
trainer.extend(extensions.ProgressBar())
# Save vocabulary and model's setting
if not os.path.isdir(self.out):
os.mkdir(self.out)
vocab_path = os.path.join(self.out, 'vocab.json')
with open(vocab_path, 'w') as f:
json.dump(vocab, f)
model_path = os.path.join(self.out, 'best_model.npz')
experiment_setup = self.__dict__
# TODO: move all this to the parent class
experiment_setup['task'] = "text classification"
experiment_setup['vocab_path'] = vocab_path
experiment_setup['model_path'] = model_path
experiment_setup['n_class'] = n_class
experiment_setup['datetime'] = self.current_datetime
with open(os.path.join(self.out, 'args.json'), 'w') as f:
json.dump(self.__dict__, f)
# Run the training
trainer.run()
result = {}
result['experiment_setup'] = experiment_setup
result['experiment_setup']['default_measurement'] = 'accuracy'
result['experiment_setup']['dataset'] = os.path.basename(
os.path.normpath(path_dataset))
result['experiment_setup']['method'] = self.model
result['experiment_setup']['embeddings'] = embeddings.metadata
result['log'] = load_json(os.path.join(self.out, 'log'))
# TODO: old version was returning last test value, make a footnote
# result['result'] = {"accuracy": result['log'][-1]['validation/main/accuracy']}
accuracy = max(_["validation/main/accuracy"] for _ in result['log'])
result['result'] = {"accuracy": accuracy}
return [result]
