Skip to content

Latest commit

 

History

History

stable-baselines

README.md

stable-baselines 腳本

build_docker.sh

我們可以建置４種版本的image分別是(GPU, GPU-MPI, CPU, CPU-MPI) 通常只有使用(GAIL, DDPG, TRPO, PPO1)需要安裝MPI。因為openMPI有些問題，建議使用沒有MPI的版本。 詳細請參考: stable-baselines 安裝

GPU

USE_GPU=True ./scripts/build_docker.sh

GPU-MPI

USE_GPU=True INSTALL_MPI=True ./scripts/build_docker.sh

CPU

./scripts/build_docker.sh

CPU-MPI

INSTALL_MPI=True ./scripts/build_docker.sh

run_docker

假如:

stable-baselines 專案目錄:
/homes/user/stable-baselines/

運行 GPU 版本

Build and run Docker containers leveraging NVIDIA GPUs

# Default all GPU
SB_PATH=/homes/user/stable-baselines/ ./run_docker_gpu.sh python train.py

# Starting a GPU enabled container on specific GPUs
# https://github.com/NVIDIA/nvidia-docker#usage
GPUS="device=1,2" SB_PATH=/homes/user/stable-baselines/ ./run_docker_gpu.sh python train.py

運行 CPU 版本

SB_PATH=/homes/user/stable-baselines/ ./run_docker_cpu.sh python train.py

run_docker_tensorboard

移動到 tensorboard 紀錄存放資料夾

# Default port: 6006
./run_docker_tensorboard.sh tensorboard --logdir=./ --host=0.0.0.0

# Port 80
TARGET_PORT=80 ./run_docker_tensorboard.sh tensorboard --logdir=./ --host=0.0.0.0