templates

Templates

Templates for agent development.

Check out the package documentation to learn more about the agent and competition framework.

Available templates

• basic.py this lets you run a preconfigured baseline agent without further development
• advanced.py this lets you configure the strategy of the baseline agent without the need to implement the whole agent.
• expert.py this lets you start from scratch, only borrowing our agent architecture which separates the agent loop from the event loop.

Testing with Sandbox (recommended)

To test your agent run it against baseline agents in the sandbox. Follow the steps 1.-3. in sandbox readme, then start your own agent:

  python templates/v1/basic.py --name basic1 --dashboard

The following additional parameters can be used to tune the agent:

• --register-as: choices=['seller', 'buyer', 'both'], The string indicates whether the baseline agent registers as seller, buyer or both on the oef.
• --search-for, choices=['sellers', 'buyers', 'both'], The string indicates whether the baseline agent searches for sellers, buyers or both on the oef.
• --is-world-modeling, type=bool, Whether the agent uses a workd model or not.
• --services-interval, type=int, The number of seconds to wait before doing another search.

Generate private key (optional)

A private key for your agent is generated every time by the library. If you want to use the same cryptographic key, you can follow these steps:

• Generate a private key:

  python scripts/generate_private_key.py private_key.pem
  

• Every time you run your agent, add the parameter --private-key-pem <pem-file> to your command:

  python templates/v1/basic.py --name basic1 --dashboard --private-key-pem private_key.pem
  

Testing manually (not recommended)

• First, start the oef:

  python scripts/oef/launch.py -c ./scripts/oef/launch_config.json
  

• Second, start the visdom server in shell:

  python -m visdom.server
  

• Third, tart the controller, followed by two agents in separate terminals.

  python tac/platform/controller.py --verbose --registration-timeout 20 --nb-agents 2 --tx-fee 0.0 --dashboard
  
  python templates/v1/basic.py --name basic0 --dashboard
  
  templates/v1/basic.py --name basic1 --dashboard
  

The following parameters can be used to tune the agent:

• --register-as: choices=['seller', 'buyer', 'both'], The string indicates whether the baseline agent registers as seller, buyer or both on the oef.

• --search-for, choices=['sellers', 'buyers', 'both'], The string indicates whether the baseline agent searches for sellers, buyers or both on the oef.

• --is-world-modeling, type=bool, Whether the agent uses a workd model or not.

• --services-interval, type=int, The number of seconds to wait before doing another search.

• Fourth, navigate to http://localhost:8097/ and select the appropriate environment.