def main():
global_config.assign_config()
if args['device'] == "cpu":
os.environ['CUDA_VISIBLE_DEVICES'] = ''
else:
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
rm_file("./data/log/*log.txt")
if args['mode'] == "train":
Train().run()
elif args['mode'] == "new_train":
print("Clean chceck point: train_dir ")
clean_folder(global_config.global_config.train_dir)
print("Clean log directory")
clean_folder(global_config.global_config.log_dir)
Train().run()
elif args['mode'] == "eval":
Evaluate().run()
if path not in sys.path:
sys.path.insert(0, path)
this_dir = osp.dirname(__file__)
package_path = osp.join(this_dir, '..')
add_path(package_path)
import tensorflow as tf
import numpy as np
from pprint import pprint as pp
from graph.forward.Iforward import IForward
from utils.config import global_config
global_config.assign_config()
iforward = IForward('train', None)
# define variable
logit_input = tf.placeholder(shape=[None, None], dtype=tf.float32)
label_input = tf.placeholder(shape=[None], dtype=tf.int64)
# build graph
def get_class_softmax_loss(logit_input, label_input):
'''
tf.nn.sparse_softmax_cross_entropy_with_logits
will apply: softmax to logit_input, one_hot encoded to label_input
then compute cross-entropy between (above) two value to get losses
def run(self):
# set evn to use CPU only
global_config.assign_config()
self.build_computation_graph()
def run(self):
global_config.assign_config()
# self.prepared_dataset = PrepareDataset()
self.build_computation_graph()
self.run_training()