Environment Execution

An environment episode can be manually executed using it’s reset() and step() methods. See the simple example below that runs through a single complete episode, using random actions sampled from the agent’s action spaces and prints out the evaluated actions and observations:

env = SupplyChainEnv()

observations = env.reset()

while not env.is_terminated() or env.is_truncated():
    actions = {
        agent.id: agent.action_space.sample()
        for agent in env.strategic_agents
    }

    observations = env.step(actions).observations

If this method is used, any supertypes must manually be passed to the agent’s __init__() method or manually set as the supertype property of the agent before the reset() method is called:

env = SupplyChainEnv()

env["SHOP"].supertype = ShopAgent.Supertype(...)

Enabling Telemetry

Phantom includes a powerful tool for aiding the development and debugging of environments called Telemetry. There are two output destinations for the output: print logging (to the terminal) and file logging:

ph.telemetry.logger.configure_print_logging(enable=True)
ph.telemetry.logger.configure_file_logging(file_path="log.json", append=False)

env = SupplyChainEnv()

observations = env.reset()

while not env.is_terminated() or env.is_truncated():
    actions = {
        agent.id: agent.action_space.sample()
        for agent in env.strategic_agents
    }

    observations = env.step(actions).observations

There are many options to configure what will be logged for both functions, see the Telemetry Logger page for full details.

Warning

This feature is only designed for single-process execution. Using this with multiple processes/workers may lead to invalid output.

The following is an example of what the print logging looks like:

Phantom also comes with a handy easy to use web viewer for log files:

This can be opened by running:

streamlit run scripts/view_telemetry.py <log_file>

Once started, navigate to http://localhost:8501 in your browser.