GitXplorerGitXplorer
M

baselines-experiments

public
5 stars
1 forks
0 issues

Commits

List of commits on branch master.
Unverified
5d899e0d2f86ae46c29a7644b24cd52d66cba9ae

Inspect results per-bucket.

MMillionIntegrals committed 6 years ago
Unverified
bb57877326574c768f0f777ea5c74878abdc615f

Local mujoco scripts.

MMillionIntegrals committed 6 years ago
Unverified
51c57577352eb626d5c2b91079515a1f445ac2da

MuJoCo experiments.

MMillionIntegrals committed 6 years ago
Unverified
cdea4cb2c3cdc32c50250641c1a8cb24507821e0

PPO MuJoCo version updates.

MMillionIntegrals committed 6 years ago
Unverified
fdf21e7303ac89504fb77f328b1dfa19aa16bfd5

Updated PPO configuration files.

MMillionIntegrals committed 6 years ago
Unverified
26e24662866c364f984d04ec8570c5a28c5e51ac

Added extra info about MongoDB to the readme.

MMillionIntegrals committed 6 years ago

README

The README file for this repository.

baselines-experiments

Infra setup to run reinforcement learning baselines in a cloud deployment.

Steps

  1. Create GCP instances to run the experiments
bash ./scripts/gcp-create-instances.sh
  1. Set up ansible inventory file. Example for GCP below
plugin: gcp_compute
projects:
  - projectname
auth_kind: serviceaccount
groups:
  rl: true
filters:
  - 'labels.algo = ppo'
  - any filters you may want
service_account_file: /home/yourdir/gcp/serviceaccount.json
  1. Install ansible galaxy roles
ansible-galaxy install geerlingguy.docker geerlingguy.pip
  1. Change ansible vault file to contain proper MongoDB login information for your database

  2. Set up experiment environment on all machines

bash ./scripts/ansible-full-setup.sh
  1. Start running the experiments
bash ./scripts/ansible-run-experiments.sh