To train a model, run
python run_trainer.py \
--data_path=/path/to/cifar10/binary/files \
Note that you can terminate training prematurely, and pick up where you left off by setting --ckpt_path=
to the path to the directory containing all checkpoint files generated so far. The parameters will be restored from the most recent checkpoint. Also, the training metrics (loss and accuracy) will be written to ./log
. Run tensorboard --logdir=log
to view tensorboard.
To evaluate a model, run
python run_evaluator.py \
--data_path=/path/to/cifar10/binary/files \
--ckpt_path=/path/to/directory/ckpt/files/will/be/loaded/from
To see full list of arguments, run
python run_trainer.py --help
python run_evaluator.py --help