./scripts/glearn train space_invaders
./scripts/glearn train space_invaders_cnn
./scripts/glearn train mnist
./scripts/glearn train mnist_vae
./scripts/glearn train mnist_cvae
./scripts/glearn train mnist_gan
./scripts/glearn train mnist_vae_gan
./scripts/glearn train mnist_dcgan
./scripts/glearn train cifar10
./scripts/glearn train digit_repeat
./scripts/glearn train ptb
./scripts/glearn train mountain_car_dqn
./scripts/glearn train mountain_car_a2c
./scripts/glearn train mountain_car_sac
Rendering is disabled during training by default, and enabled by default during evaluation. You can enable rendering with: ./scripts/glearn train ... --render
You can enable debugging information with: ./scripts/glearn train ... --debug
There are also various debug options that can be enabled in the experiment config.
The _base.yaml
file has most of these listed, which can be imported into another config.
This environment variable can be used to log TF memory info:
TF_CPP_MIN_VLOG_LEVEL
=3
./scripts/glearn train ... --version=VERSION VERSION is which params to load. It can be an int or string. If int, then autosaves to same VERSION.
./scripts/glearn train ... --profile OR python utils/profile.py --path=/path/to/profile_100