Skip to content

Latest commit

 

History

History
185 lines (121 loc) · 6.46 KB

Guidelines-Docker-GVGAI-RLbaselines.md

File metadata and controls

185 lines (121 loc) · 6.46 KB

Guidelines-Docker-GVGAI-RLbaselines.md

Step-by-step instructions written by Hao Tong.

1. Install docker

2. Download required packages

Only For GPU users

If you have GPU in your machine, you can use GPU to speed up the learning process. Before doing, you will need to:

  • install NAVIDA driver on your machine

  • install the docker on your machine

  • install NAVIDA docker (It is simply a plugin to Docker to support GPU using in container)

The above three step can be refer to the existing blog by Trung Tran: Installing NVIDIA Docker On Ubuntu 16.04

3. Dockerfile (modified from the rl_baseline repository)

CPU only (download Dockerfile)

FROM ubuntu:16.04

# Download jdk-9 from  https://www.oracle.com/technetwork/java/javase/downloads/java-archive-javase9-3934878.html
# You can replace ``jdk-9.tar.gz'' by the one that you have downloaded
ADD jdk-9.tar.gz /usr/local

RUN apt-get -y update && apt-get -y install vim git libopenmpi-dev zlib1g-dev cmake libglib2.0-0 libsm6 libxext6 libfontconfig1 libxrender1

WORKDIR /app

# Download Anaconda from https://www.anaconda.com/distribution/ 
# You can replace ``anaconda.sh'' by the installation script that you have downloaded, e.g., ``Anaconda3-5.3.1-Linux-x86_64.sh''
COPY anaconda.sh /app
RUN bash anaconda.sh -b -p /opt/conda && \
    rm -rf anaconda.sh


ENV JAVA_HOME=/usr/local/jdk-9.0.4
ENV CLASSPATH=.:$JAVA_HOME/lib:$CLASSPATH
ENV PATH=$JAVA_HOME/bin:$PATH
ENV PATH=/opt/conda/bin:$PATH

RUN git clone https://github.com/SUSTechGameAI/GVGAI_GYM.git

# The rl-baseline required ternsorflow version to be from 1.8.0 to 1.14.0
RUN pip install -e GVGAI_GYM/ && \
    pip install codacy-coverage && \
    pip install mpi4py && \
    pip install tensorflow==1.14.0 && \
    pip install cloudpickle && \
    pip install stable-baselines[mpi]

# Install pytorch cpu version by uncommenting the following line   
# RUN pip install torch==1.3.0+cpu torchvision==0.4.1+cpu -f https://download.pytorch.org/whl/torch_stable.html

CMD /bin/bash

GPU version (download Dockerfile)

  • In the dockerfile, use FROM nvidia/cuda:9.0-cudnn7-runtime-ubuntu16.04 instead of From Ubuntu:16.04.

  • The version of TensorFlow and Pytorch should be GPU version.

  • When run container from image, add --runtime=nvdia in the command.

    E.g. docker run --runtime=nvidia -v $PWD:/home --rm -it baseline_gpu /bin/bash

  • Use nvidia-smi to find out the status of GPU.

FROM nvidia/cuda:9.0-cudnn7-runtime-ubuntu16.04

# Download jdk-9 from  https://www.oracle.com/technetwork/java/javase/downloads/java-archive-javase9-3934878.html
# You can replace ``jdk-9.tar.gz'' by the one that you have downloaded
ADD jdk-9.tar.gz /usr/local

RUN apt-get -y update && apt-get -y install vim git libopenmpi-dev zlib1g-dev cmake libglib2.0-0 libsm6 libxext6 libfontconfig1 libxrender1

WORKDIR /app

# Download Anaconda from https://www.anaconda.com/distribution/ 
# You can replace ``anaconda.sh'' by the installation script that you have downloaded, e.g., ``Anaconda3-5.3.1-Linux-x86_64.sh''
COPY anaconda.sh /app
RUN bash anaconda.sh -b -p /opt/conda && \
    rm -rf anaconda.sh


ENV JAVA_HOME=/usr/local/jdk-9.0.4
ENV CLASSPATH=.:$JAVA_HOME/lib:$CLASSPATH
ENV PATH=$JAVA_HOME/bin:$PATH
ENV PATH=/opt/conda/bin:$PATH

RUN git clone https://github.com/SUSTechGameAI/GVGAI_GYM.git

# The rl-baseline required ternsorflow version to be from 1.8.0 to 1.14.0
RUN pip install -e GVGAI_GYM/ && \
    pip install codacy-coverage && \
    pip install mpi4py && \
    pip install tensorflow-gpu==1.14 && \
    pip install cloudpickle && \
    pip install stable-baselines[mpi]

# Install pytorch cpu version by uncommenting the following line   
# RUN conda install pytorch==1.1.0 torchvision==0.3.0 cudatoolkit==9.0 -c pytorch

CMD /bin/bash

Remark: anaconda.sh jdk.tar.gz and Dockerfile should be in the same folder!

4. Build your image

Now you should have the following files in the same repository:

  • jdk-9.tar.gz
  • anaconda.sh
  • Dockfile

Try the follwing command line in the repository:

docker build . -t <image_name>

5. Testing the installation

  • Use docker ps to check whether your image exist or not.
  • Tests the environment by using pytest runner:

    • CPU version:

      docker run -w /app/GVGAI_GYM --rm <image_name> pytest

    • GPU version:

      docker run --runtime=nvidia -w /app/GVGAI_GYM --rm <image_name> pytest

6. Run container from image

CPU only

docker run -v $PWD:/home -w /home --name <container_name> -it <image_name> /bin/bash

For GPU users

docker run --runtime=nvidia -v $PWD:/home -w /home --name <container_name> -it <image_name> /bin/bash

ps: The -v flag mounts the current working directory into the container.

Other commands

  • Exit a running container: CTRL + D

  • Load an existing container:

docker start <container_name>

  • Run your docker in the background add -d before -it

  • Enter into a container running in the background

docker exec -it <contanier_name> /bin/sh

More docker commands refer to the official tutorial

References

  1. G. Brockman, V. Cheung, L. Pettersson, J. Schneider, J. Schulman, J. Tang, and W. Zaremba, “Openai gym,” 2016.
  2. A. Hill, A. Raffin, M. Ernestus, A. Gleave, A. Kanervisto, R. Traore, P. Dhariwal, C. Hesse, O. Klimov, A. Nichol, M. Plappert, A. Radford, J. Schulman, S. Sidor, and Y. Wu, “Stable baselines,” https://github.com/hill-a/stable-baselines, 2018.
  3. R. R. Torrado, P. Bontrager, J. Togelius, J. Liu, and D. Perez-Liebana, “Deep reinforcement learning for general video game ai,” in Computational Intelligence and Games (CIG), 2018 IEEE Conference on. IEEE, 2018.