def get_data(train_portion=0.7):
''' put the protein data into bucketes and split into train, dev, test sets
Args:
train_portion(default=0.7): the ratio of training data for splitting
Returns:
train_set, dev_set, test_set: data sets paired and put in buckets
'''
print("Reading data...")
raw_data = dr.load_data()
print(np.array(raw_data).shape)
global feature_size
feature_size = len(raw_data[0][0])
pair_data = []
for i in xrange(len(raw_data)):
pair = (raw_data[i],raw_data[i])
pair_data.append(pair)
train_data, test_data =dr.split_train_test(pair_data,train_portion)
train_set = (train_data,) # only one bucket
test_set = (test_data,)
dev_set = (test_data,) # for simplicity
# print(np.array(train_set).shape)
return train_set,dev_set,test_set
def main(unused_args):
# if not FLAGS.data_path:
# raise ValueError("Must set --data_path to PTB data directory")
config = SmallConfig()
with tf.Graph().as_default(), tf.Session() as session:
raw_data = np.array(datareader.load_data())
# print raw_data
config.feature_size = len(raw_data[0, 0, :])
# print config.feature_size
initializer = tf.random_uniform_initializer(-config.init_scale, config.init_scale)
with tf.variable_scope("model", reuse=None, initializer=initializer):
m = LSTMModel(True, config)
tf.initialize_all_variables().run()
print "done"