This page lays out some common design patterns for setting up your Phantom environment.
This pattern is common for simple experiments. In this pattern all agents act at every step. Subclassing an environment from the :class:`PhantomEnv` class is sufficient for this.
For more complex experiments with multiple steps and multiple agent groups potentially with a subset of agents taking actions each turn Phantom provides the :class:`FiniteStateMachineEnv` class. This formalises the environment as a free state machine by extending the :class:`PhantomEnv` class and providing a clean interface for managing the logic of each state and the transitions between states.
A simple but commonly used pattern is that of a free state machine with two states that pass from one another. This is formally known as a Stackleberg game or a Stackelberg competition. This can easily be implemented using the :class:`FiniteStateMachineEnv` class.
