chenjiaqiang-a/generative-deep-learning

复现generative-deep-learning中的示例

generative-deep-learning

复现generative-deep-learning中的示例

有用的script

jupyter nbconvert --to script test.ipynb

nohup python test.py &

ps -aux | grep "test.py"

kill -9 PID

zip -r test.zip ./test

世界模型 WorldModel

• 收集随机 rollout 数据

python ./world_model/generate_data.py car_racing --total_episodes 2000 --time_steps 300

• 训练VAE

python ./world_model/train_vae.py --new_model

• 收集用于训练RNN的数据

python ./world_model/generate_rnn_data.py

• 训练RNN

python ./world_model/train_rnn.py --new_model

• 训练Controller

# subprocess.py > class Popen > __init__ > shell=True

python ./world_model/train_controller.py car_racing --num_worker 2 --num_worker_trial 2 --num_episode 4 --max_length 1000 --eval_steps 25

python ./world_model/train_controller.py car_racing --num_worker 2 --num_worker_trial 2 --num_episode 4 --max_length 1000 --eval_steps 25 --dream_mode 1

• 查看训练结果

python ./world_model/base/model.py --filename [json filename] --render_mode --record_video