Skip to content

Latest commit

 

History

History
62 lines (35 loc) · 2.47 KB

aws_readme.md

File metadata and controls

62 lines (35 loc) · 2.47 KB

AWS Readme

These are instructions for running scoping experiments on aws EC2. Note that the results in the paper are from experiments run on a laptop using a different set of scripts, NOT in EC2 using the scripts described here. You can find a tutorial on using AWS EC2 here.

Instance parameters

These instructions were verified to work using an instance with the following configuration:

  • Type: t2.large

  • AMI

    • AMI ID: ami-0ac5a3737d540e82a
    • AMI Name: Deep Learning Base AMI (Ubuntu 16.04) Version 40.0
    • AMI Location: amazon/Deep Learning Base AMI (Ubuntu 16.04) Version 40.0

  • Network

Usage

Copy the scoping_supplement.zip to your ec2 instance with scp:

scp -i <path2key_file> scoping_supplement.zip ubuntu@<ec2_instance_address>:~/

SSH into the ec2 instance

ssh -i <path2key_file> ubuntu@<ec2_instance_address>

Unzip the supplement

unzip scoping_supplement.zip

Mount the planutils docker image

source scoping_supplement/scoping/ubuntu_scripts/get_enhsp_docker.sh

Change directory to /scoping_supplement

cd /scoping_supplement

Run config_all.sh to setup conda, fast downward

source scoping/ubuntu_scripts/config_all.sh

Run ENHSP-2020 experiments using the experiments/enhsp_experiment.py script. The arguments are:

  1. Number of runs in this experiment
  2. Path to domain file
  3. Path to problem file
  4. Path to directory the experiment logs will be saved in
  5. (optional) --force_clear_log_dir if you want to forcibly clear the passed experiment log directory. If you pass an already existing directory as the experiment log directory and don't use the --force_clear_log_dir option, the script will raise an exception.

Example command:

python ./experiments/enhsp_experiment.py 2 examples/multi_monkeys_playroom/multi_monkeys_playroom.pddl examples/multi_monkeys_playroom/prob01.pddl ./experiment_logs/multi_monkeys_playroom_01

Run Fast Downward experiments using the experiments/fd_experiment.py script. The arguments are the same as for the enhsp script. Example command:

python experiments/fd_experiment.py 1 examples/IPC_domains_propositional/logistics00/domain.pddl examples/IPC_domains_propositional/logistics00/prob15.pddl ../downward/fast-downward.py experiment_logs/logistics15

To move results out of the ec2 isntance, you can use scp from your local machine:

scp -i <path2key_file> ubuntu@<ec2_instance_address>:<path2experiments_log_dir> scoping_experiment_logs