python inc_stop.py
python dup_add.py
python dup_add.py --submask
python reg_jmp.py
python sub.py
python sub.py --sketch --hard --sketch_no_jmp
python sub.py --train_data_with_branch --sketch --hard
python sub.py --train_data_with_branch --sketch --soft
python sub.py --train_data_with_sub --sketch --hard
python sub.py --train_data_with_sub --sketch --soft
python sub.py --train_data_with_sub --sketch --soft --sketch_no_jmp --training_steps 100000
python sub.py --hard --sketch --mask_pc --mask_data_p --mask_data --mask_ret_p mask_ret
python sub.py --soft --sketch --mask_pc --mask_data_p --mask_data --mask_ret_p mask_ret
python sub.py --gumbel_softmax
python sub.py --train_data_with_branch
python sub.py --sketch --hard --sketch_no_jmp --training_steps 500000
(though it worked without the sketch, and though it worked with fewer training steps!)python sub.py --train_data_with_sub --training_steps 5000000
python sub.py --train_data_with_sub --sketch --hard --sketch_no_jmp --training_steps 500000