README.md

This is a project for the poker experiments in the paper Actor-Critic Policy Optimization in a Large-Scale Imperfect-Information Game.

This project depends on a customized version of Open_spiel (See third_party/open_spiel). Other dependencies include Tensorflow and project_acpc_server.

Requirements.

Manually installation.

1. install required packages.

   apt update && apt install ssh git build-essential curl wget python3 python3-dev python3-pip python3-setuptools python3-wheel python3-tk autoconf automake libtool libffi-dev
   # install cmake-3.18
   wget -qO - https://apt.kitware.com/keys/kitware-archive-latest.asc | apt-key add -
   apt-add-repository 'deb https://apt.kitware.com/ubuntu/ bionic main'
   apt update && apt install cmake

2. install gperftools

   # install libunwind-1.1
   wget http://download.savannah.gnu.org/releases/libunwind/libunwind-1.1.tar.gz 
   tar -xzf libunwind-1.1.tar.gz 
   cd libunwind-1.1    
   ./configure  
   make -j8
   make install 
   # install gperftools
   wget https://github.com/gperftools/gperftools/releases/download/gperftools-2.8/gperftools-2.8.tar.gz
   tar -xzf gperftools-2.8.tar.gz
   cd gperftools-2.8 
   ./configure
   make -j8
   make install

3. install tensorflow-cpp

   # apt install
   apt install wget unzip autoconf automake libtool zlib1g-dev liblzma-dev
   # install bazel-0.26.1
   wget https://github.com/bazelbuild/bazel/releases/download/0.26.1/bazel-0.26.1-installer-linux-x86_64.sh
   chmod +x./bazel-0.26.1-installer-linux-x86_64.sh 
   ./bazel-0.26.1-installer-linux-x86_64.sh
   # clone tensorflow and compile tensorflow-cpp-lib
   git clone --branch v1.15.3  https://github.com/tensorflow/tensorflow
   cd tensorflow
   ./configure
   bazel build --config=opt --config=cuda --config=monolithic  //tensorflow:libtensorflow_cc.so

4. Copy tensorflow head files and binaries to third_party.

   cd third_party && mkdir tensorflow && cd tensorflow
   mkdir include
   cp -r /tensorflow/tensorflow ./include
   cp -r /tensorflow/third_party ./include
   cp -r /tensorflow/bazel-tensorflow/external/nsync ./include
   cp -r /tensorflow/bazel-tensorflow/external/eigen_archive ./include
   cp -r /tensorflow/bazel-tensorflow/external/com_google_absl ./include
   cp -r /tensorflow/bazel-tensorflow/external/com_google_protobuf ./include
   mkdir bazel_include
   rsync -avm --include='*.h'  -f 'hide,! */' /tensorflow/bazel-bin/ ./bazel_include
   mkdir bin
   cp -r /tensorflow/bazel-bin/tensorflow/libtensorflow_cc* ./bin

5. build algorithms.

   source env.sh
   cd third_party && ./install.sh
   cd ..
   mkdir build && cd build
   cmake -DCMAKE_BUILD_TYPE=Release ..
   make -j8

Dockerfiles.

One can also build a docker image using docker_build.sh.

How to run.

1. Make dirs for results.

mdkir results
mkdir models

2. call ach.sh, a2c.sh, neurd.sh, or rpg.sh to run the algorithm.