example-sweep.yaml

# Use this file to run a parameter sweep on wandb.
#
# First, initialize the sweep by running
#
#     wandb sweep training/example-sweep.yaml
#
# This will print out a sweep id. Then just run
#
#     wandb agent <sweep-id>
#
# or
#
#     gcloud/start-remote-job <remote> <sweep-id> --sweep
#
# and that should be it! You can start as many concurrent agents as you want.
# More details can be found at https://docs.wandb.com/sweeps/quickstart.

---

program: start-training.py
method: grid
command:
  - python3
  - start-training.py
  - --wandb

parameters:

  env_type:
    # See training/env_factory.py for other values
    values:
      - append-spawn
      - prune-spawn

  steps:
    value: 6_000_000

  algo:
    value: ppo

  side_effect.penalty:
    values: [0.0, 0.1, 0.2, 0.3, 0.4, 0.5]

  side_effect.baseline:
    value: 'starting-state'  # or 'inaction'