f['x_valid'] = x_valid, f['y_valid'] = y_valid
f.flush()
f.close()
else:
f = h5py.File(Configure.dataset_path, 'r')
x_train = f['x_train'][:]
y_train = f['y_train'][:]
x_test = f['x_test'][:]
y_test = f['y_test'][:]
x_valid = f['x_valid'][:]
y_valid = f['y_valid'][:]
f.close()
print("train/valid/test split: {:d}/{:d}/{:d}".format(len(y_train), len(y_valid), len(y_test)))
print('---> create train/valid data wapper')
train_data_wrapper = data_util.DataWrapper(x_train, y_train, istrain=True)
valid_data_wrapper = data_util.DataWrapper(x_valid, y_valid, istrain=True)
test_data_wrapper = data_util.DataWrapper(x_valid, y_valid, istrain=False)
print('---> build model')
# Built model and start training
# ==================================================
with tf.Graph().as_default(), tf.device('/gpu:2'):
session_conf = tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=False)
session = tf.Session(config=session_conf)
with session.as_default():
text_rcnn = TextRCNN(
label_size=FLAGS.label_size,
sequence_length=max_document_length,
print('---> load test text dataset')
# datasets = data_util.get_datasets_20newsgroup()
# x_text, y = data_util.load_data_labels(datasets)
x_text, y = data_util.load_text_datasets()
# Restore vocabulary in checkpoint dir
vocabulary_path = os.path.join(FLAGS.checkpoint_dir, '..', 'vocabulary')
vocab_processor = learn.preprocessing.VocabularyProcessor.restore(
filename=vocabulary_path)
# map text to vocabulary index
x_text = np.array(list(vocab_processor.transform(x_text)))
print('---> create test data wapper')
test_data_wrapper = data_util.DataWrapper(x_text,
istrain=False,
is_shuffle=False)
# Evaluation
# ==================================================
checkpoint_file = tf.train.latest_checkpoint(FLAGS.checkpoint_dir)
print(checkpoint_file)
with tf.Graph().as_default(), tf.device('/gpu:2'):
session_conf = tf.ConfigProto(allow_soft_placement=True,
log_device_placement=False)
session = tf.Session(config=session_conf)
with session.as_default():
# Load the saved meta graph and restore variables
saver = tf.train.import_meta_graph("{}.meta".format(checkpoint_file))
saver.restore(session, checkpoint_file)