Skip to content

Latest commit

 

History

History
72 lines (54 loc) · 2.41 KB

run_in_aws.md

File metadata and controls

72 lines (54 loc) · 2.41 KB
Raw

Training can also be run in AWS to take advantage of cloud GPUs. Launch Templates are used to automate a lot of the instance setup.

Step 1: Launch Spot Instance

  • Use Launch Template to start instance (see below for setup)

Step 2: On Every SSH

source activate pytorch_p36
export DISPLAY=:0

Launch Template

  • Region: us-east-1
  • AMI ID: ami-18642967
  • Instance Type: p2.xlarge
  • Security Group: ssh-inbound-only (you'll need to create your own)
  • Key Pair: daniel (you'll need to create your own)
  • Purchasing option: spot
  • User Data (see below)

User Data Script

#!/bin/bash

/usr/bin/X :0 &
su - -c "conda install -y -n pytorch_p36 pip swig opencv scikit-image" ubuntu
su - -c "source activate pytorch_p36; pip install torchsummary tensorboardX dill gym Box2D box2d-py unityagents pygame ppaquette_gym_super_mario" ubuntu
DIR="/home/ubuntu"

# install gym toolkit
su -c "cd $DIR; git clone https://github.com/openai/gym.git" ubuntu
su - -c "source activate pytorch_p36; cd $DIR/gym; pip install -e '.[atari]'" ubuntu

# install ml agents toolkit
su -c "cd $DIR/ml-agents; git pull" ubuntu
su - -c "source activate pytorch_p36; cd $DIR/ml-agents/ml-agents; pip install ." ubuntu

# install PLE toolkit
su -c "cd $DIR; git clone https://github.com/ntasfi/PyGame-Learning-Environment" ubuntu
su - -c "source activate pytorch_p36; cd $DIR/PyGame-Learning-Environment; pip install -e ." ubuntu

# install angela
cd $DIR
su - -c "git clone https://github.com/danielnbarbosa/angela.git" ubuntu

# install udacity unity ml environments
cd $DIR/angela/compiled_unity_environments
wget https://s3-us-west-1.amazonaws.com/udacity-drlnd/P1/Banana/VisualBanana_Linux.zip
wget https://s3-us-west-1.amazonaws.com/udacity-drlnd/P2/Reacher/Reacher_Linux.zip
wget https://s3-us-west-1.amazonaws.com/udacity-drlnd/P2/Crawler/Crawler_Linux.zip
wget https://s3-us-west-1.amazonaws.com/udacity-drlnd/P3/Tennis/Tennis_Linux.zip
wget https://s3-us-west-1.amazonaws.com/udacity-drlnd/P3/Soccer/Soccer_Linux.zip
unzip VisualBanana_Linux.zip
unzip Reacher_Linux.zip
unzip Crawler_Linux.zip
unzip Tennis_Linux.zip
unzip Soccer_Linux.zip
rm VisualBanana_Linux.zip Reacher_Linux.zip Crawler_Linux.zip Tennis_Linux.zip Soccer_Linux.zip
chown -R ubuntu:ubuntu VisualBanana_Linux Reacher_Linux Crawler_Linux Tennis_Linux Soccer_Linux

# update packages
su - -c "conda update -y -n pytorch_p36 --all" ubuntu

# install NES emulator
apt-get -y install fceux