def do_train(args):
pretrained_embeddings, token2id = util.load_word_embedding(input_file=args.vectors, cache='cache')
stopwords = util.load_stopwords()
stopwords = None
train_data = util.Data(args.data_train, args.ltp_data, stopwords=stopwords)
dev_data = util.Data(args.data_dev, args.ltp_data, max_length=train_data.max_length, stopwords=stopwords)
config = Config(args)
print(train_data.max_length)
# 配置参数. 测试集如何设置?
_, config.max_length = train_data.get_metadata()
config.n_classes = len(train_data.LABELS)
config.n_word_embed_size = len(pretrained_embeddings[0])
with tf.Graph().as_default():
logger.info("Building model...",)
start = time.time()
model = Classifier(pretrained_embeddings, token2id, config)
logger.info("took %.2f seconds", time.time() - start)
init = tf.global_variables_initializer()
saver = tf.train.Saver()
gpu_options = tf.GPUOptions(allow_growth=True)
with tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) as session:
session.run(init)
score = model.fit(session, saver, train_data, dev_data)
print("\n")
logger.info("training finished, took %.2f seconds with P: %.2f", time.time() - start, score)
def do_predict(args):
pretrained_embeddings, token2id = util.load_word_embedding(input_file=args.vectors, cache='cache')
stopwords = util.load_stopwords()
stopwords = None
train_data = util.Data(args.data_train, args.ltp_data, stopwords=stopwords)
test_data = util.Data(args.data_test, args.ltp_data, max_length=train_data.max_length, stopwords=stopwords)
config = Config(args)
# 配置参数. 测试集如何设置?
_, config.max_length = train_data.get_metadata()
config.n_classes = len(train_data.LABELS)
config.n_word_embed_size = len(pretrained_embeddings[0])
config.batch_size = len(test_data.data)
with tf.Graph().as_default():
logger.info("Building model...",)
start = time.time()
model = Classifier(pretrained_embeddings, token2id, config)
logger.info("took %.2f seconds", time.time() - start)
init = tf.global_variables_initializer()
saver = tf.train.Saver()
gpu_options = tf.GPUOptions(allow_growth=True)
with tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) as session:
session.run(init)
saver.restore(session, model.config.output_model)
labels, prediction = model.output(session, test_data, None)
print(labels)
print(prediction)
test_data.update_labels(prediction).save_result()
def svhn_test(model, num_label=500):
# Load SVHN dataset
# load data on the cpu
with tf.device('/CPU:0'):
X_train, y_train = tfds.as_numpy(
tfds.load('svhn_cropped',
split='train',
as_supervised=True,
batch_size=-1))
X_test, y_test = tfds.as_numpy(
tfds.load('svhn_cropped',
split='test',
as_supervised=True,
batch_size=-1))
# one-hot encode the outs
y_train = np.eye(10)[y_train.reshape(-1)]
y_test = np.eye(10)[y_test.reshape(-1)]
# cast it all to floats for image augmentation, rescale to [0,1]
X_train = X_train.astype('float32') / np.float(255.0)
X_test = X_test.astype('float32') / np.float(255.0)
# whiten the data or apply zca
X_train = whiten_norm(X_train)
X_test = whiten_norm(X_test)
print('loaded svfn', X_train.shape, X_test.shape)
# Setup test set
test = util.Data(X_test, y_test, None)
# Split training test into labeled and unlabeled
train = util.label_unlabel_split(X_train, y_train, num_label, 10)
# Split training data into training and validation
(train, valid) = util.train_test_valid_split(train.X,
train.y,
split=(0.9, 0.1),
U=train.U)
print('TR:', train.X.shape, train.y.shape, train.U.shape)
print('v', valid.X.shape, valid.y.shape)
# fit on the gpu
with tf.device('/GPU:0'):
# Train model using training and validation sets
hist = model.fit(train, valid)
print('evaluating on (subset) of test set...')
with tf.device('/CPU:0'):
# Test the model using test set
y_pred = model.predict(test.X[0:1000])
acc = float(
tf.reduce_mean(
tf.keras.metrics.categorical_accuracy(test.y[0:1000], y_pred)))
print(model.name, ' : acc:', acc)
return model, {'hist': hist, 'acc': acc}
def make_request(self, is_part_upload=False, part_index=None,
file_location=None, save_path_parent=None,
file_name=None, is_range_download=False,
part_download_queue=None, range_start=None,
stop_flag_obj=None, is_last_retry=False):
has_none_been_put = False
chunk_size = 65536
peer_addr = self.my_http_connection.host
local_addr = ''
http_response = None
recv_body = ''
self.response.start_time = time.time()
try:
self.my_http_connection.connection.putrequest(
self.obs_request.method, self.obs_request.url, skip_host=1)
# send headers
for k in self.obs_request.headers.keys():
if isinstance(self.obs_request.headers[k], list):
for i in self.obs_request.headers[k]:
self.my_http_connection.connection.putheader(k, i)
else:
self.my_http_connection.connection.putheader(k,
self.obs_request.headers[
k])
self.my_http_connection.connection.endheaders()
local_addr = str(
self.my_http_connection.connection.sock._sock.getsockname())
peer_addr = str(
self.my_http_connection.connection.sock._sock.getpeername())
logging.debug(
'Request:[%s], conn:[%s->%s], sendURL:[%s], sendHeaders:[%r], sendContent:[%s]' % (
self.obs_request.request_type, local_addr, peer_addr,
self.obs_request.url, self.obs_request.headers,
self.obs_request.send_content[0:1024]))
if self.obs_request.content_length > 0 and not self.obs_request.send_content:
if is_part_upload:
with open(file_location, 'rb') as obj_to_put:
obj_to_put.seek(part_index)
while self.response.send_bytes < self.obs_request.content_length:
if stop_flag_obj.flag:
raise Exception(
'Stop Because Some Part_upload Failed')
if self.obs_request.content_length - self.response.send_bytes >= chunk_size:
chunk = obj_to_put.read(chunk_size)
self.response.send_bytes += chunk_size
else:
chunk = obj_to_put.read(
self.obs_request.content_length -
self.response.send_bytes)
self.response.send_bytes += (
self.obs_request.content_length -
self.response.send_bytes)
self.my_http_connection.connection.send(chunk)
else:
with open(file_location, 'rb') as obj_to_put:
while self.response.send_bytes < self.obs_request.content_length:
if self.obs_request.content_length - self.response.send_bytes >= chunk_size:
chunk = obj_to_put.read(chunk_size)
self.response.send_bytes += chunk_size
else:
chunk = obj_to_put.read(
self.obs_request.content_length -
self.response.send_bytes)
self.response.send_bytes += (
self.obs_request.content_length -
self.response.send_bytes)
self.my_http_connection.connection.send(chunk)
else:
self.my_http_connection.connection.send(
self.obs_request.send_content)
self.response.send_bytes += len(self.obs_request.send_content)
wait_response_time_start = time.time()
logging.debug('total send bytes: %d, content-length: %d' % (
self.response.send_bytes, self.obs_request.content_length))
# get response
http_response = self.my_http_connection.connection.getresponse(
buffering=True)
wait_response_time = time.time() - wait_response_time_start
logging.debug('get response, wait time %.3f' % wait_response_time)
# read the body
content_length = int(
http_response.getheader('Content-Length', '-1'))
logging.debug('get ContentLength: %d' % content_length)
self.response.content_length = content_length
self.response.request_id = http_response.getheader(
'x-amz-request-id', '9999999999999998')
self.response.id2 = http_response.getheader('x-amz-id-2', 'None')
if http_response.status < 300 and self.obs_request.request_type == 'GetObject':
if not is_range_download:
file_path = os.path.join(save_path_parent, file_name)
save_path = os.path.dirname(file_path)
if not os.path.isdir(save_path):
try:
os.makedirs(save_path)
except:
pass
with open(file_path, 'wb') as f:
try:
while True:
chunk = http_response.read(65536)
if not chunk:
logging.info('chunk is empty, break cycle')
recv_body = '[receive content], length: %d' % self.response.recv_bytes
break
self.response.recv_bytes += len(chunk)
f.write(chunk)
except Exception, e:
logging.error('download file(%s) error(%s)' % (
self.obs_request.key, e))
try:
os.remove(file_path)
except Exception:
pass
else:
count = 0
chunk_size = 65536
while not stop_flag_obj.flag:
chunk = http_response.read(chunk_size)
if not chunk:
logging.info('chunk is empty, break cycle')
recv_body = '[receive content], length: %d' % self.response.recv_bytes
break
self.response.recv_bytes += len(chunk)
offset = range_start + chunk_size * count
data = util.Data(chunk=chunk, offset=offset)
while not stop_flag_obj.flag:
try:
part_download_queue.put(data, block=True,
timeout=1)
break
except Full:
pass
else:
logging.info(
'stop put data, range_start: %d' % range_start)
raise Exception(
'Stop Because Some Range_download Failed')
count += 1
else:
def cifar10_test(model, num_label=4000):
# load data on the cpu
with tf.device('/CPU:0'):
# Load in training and test data
X_train, y_train = tfds.as_numpy(
tfds.load('cifar10',
split='train',
as_supervised=True,
batch_size=-1)) #cifar_10.load_cifar_10()
X_test, y_test = tfds.as_numpy(
tfds.load('cifar10',
split='test',
as_supervised=True,
batch_size=-1))
# one-hot encode the outs
y_train = np.eye(10)[y_train.reshape(-1)]
# print('y_train sample:', y_train[0:10])
y_test = np.eye(10)[y_test.reshape(-1)]
# print('y_test sample:', y_test[0:10])
# cast it all to floats for image augmentation, rescale to [0,1]
X_train = X_train.astype('float32') / np.float(255.0)
X_test = X_test.astype('float32') / np.float(255.0)
# whiten the data or apply zca
X_train = whiten_norm(X_train)
X_test = whiten_norm(X_test)
# X_train = whiten_norm(X_train)
# X_test = whiten_norm(X_test)
# X_train, y_train, X_test, y_test = cifar_10.load_cifar_10()
print('loaded cifar10', X_train.shape, X_test.shape)
# Setup test set
test = util.Data(X_test, y_test, None)
# Split training test into labeled and unlabeled
train = util.label_unlabel_split(X_train, y_train, num_label, 10)
# Split training data into training and validation
(train, valid) = util.train_test_valid_split(train.X,
train.y,
split=(0.9, 0.1),
U=train.U)
# One-hot encode cifar_10.y_train and cifar_10.y_test?
## ^^ yes. Done.
print('TR:', train.X.shape, train.y.shape, train.U.shape)
print('v', valid.X.shape, valid.y.shape)
# fit on the gpu
with tf.device('/GPU:0'):
# Train model using training and validation sets
hist = model.fit(train, valid)
print('evaluating on (subset) of test set...')
with tf.device('/CPU:0'):
# Test the model using test set
y_pred = model.predict(test.X[0:1000])
# if outputs are one-hot encoded, need to decode for correctness test
# wrong = util.percent_wrong(y_pred, test.y)
# acc = 1.0 - wrong
acc = float(
tf.reduce_mean(
tf.keras.metrics.categorical_accuracy(test.y[0:1000], y_pred)))
print(model.name, ' : acc:', acc)
return model, {'hist': hist, 'acc': acc}
# TRAIN_BUFFER_SIZE = len(input_tensor_train)
TEST_BUFFER_SIZE = len(input_tensor_test)
BATCH_SIZE = 64
# TRAIN_N_BATCH = TRAIN_BUFFER_SIZE // BATCH_SIZE
TEST_N_BATCH = TEST_BUFFER_SIZE // BATCH_SIZE
embedding_dim = 256
units = 1024
vocab_inp_size = len(globalVocab.word2idx)
target_size = len(emotion_dict)
torch.cuda.set_device(0)
# train_dataset = util.Data(input_tensor_train, target_tensor_train)
test_dataset = util.Data(input_tensor_test, target_tensor_test)
# train_dataset = DataLoader(train_dataset, batch_size = BATCH_SIZE,
# drop_last=True,
# shuffle=True)
test_dataset = DataLoader(test_dataset,
batch_size=BATCH_SIZE,
drop_last=False,
shuffle=False)
# print(val_dataset.batch_size)
# device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# torch.cuda.set_device(0)
import util amazon = "amazon_cells_labelled.txt" imdb = "imdb_labelled.txt" yelp = "yelp_labelled.txt" v2 = util.Data(amazon, imdb, yelp, quiet=False) res = v2.test() print(res)
print(len(input_tensor_test))
TRAIN_BUFFER_SIZE = len(input_tensor_train)
TEST_BUFFER_SIZE = len(input_tensor_test)
BATCH_SIZE = 64
TRAIN_N_BATCH = TRAIN_BUFFER_SIZE // BATCH_SIZE
TEST_N_BATCH = TEST_BUFFER_SIZE // BATCH_SIZE
embedding_dim = 256
units = 1024
vocab_inp_size = len(globalVocab.word2idx)
target_size = len(emotion_dict)
torch.cuda.set_device(0)
train_dataset = util.Data(input_tensor_train, target_tensor_train)
test_dataset = util.Data(input_tensor_test, target_tensor_test)
train_dataset = DataLoader(train_dataset,
batch_size=BATCH_SIZE,
drop_last=True,
shuffle=True)
test_dataset = DataLoader(test_dataset,
batch_size=BATCH_SIZE,
drop_last=True,
shuffle=True)
# print(val_dataset.batch_size)
# device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
