the use of tensor2tensor
The tensorflow version should be better follow the Official document(https://github.com/tensorflow/tensor2tensor/blob/master/setup.py ),then you will meet less error
t2t-trainer \
--generate_data \
--data_dir=~/t2t_data \
--output_dir=~/t2t_train/mnist \
--problem=image_mnist \
--model=shake_shake \
--hparams_set=shake_shake_quick \
--train_steps=1000 \
--eval_steps=100
fist you should have a dir named 't2t':
haiou@gpu3-1080:~/t2t_1/t2t$ ls
decoder rawdata self_data self_script timit train
and in the dir 'self_script':
haiou@gpu3-1080:~/t2t_1/t2t$ ls self_script/
__init__.py my_problem.py
init.py:
from . import my_problem
my_problem.py: just reference to the script in:https://github.com/tensorflow/tensor2tensor/tree/master/tensor2tensor/data_generators