flatland.trajectories.policy_runner module

class flatland.trajectories.policy_runner.PolicyRunner

Bases: object

static create_from_policy(policy: Policy, data_dir: Path, env: RailEnv | None = None, n_agents=7, x_dim=30, y_dim=30, n_cities=2, max_rail_pairs_in_city=4, grid_mode=False, max_rails_between_cities=2, malfunction_duration_min=20, malfunction_duration_max=50, malfunction_interval=540, speed_ratios=None, seed=42, obs_builder: ObservationBuilder | None = None, snapshot_interval: int = 1, ep_id: str | None = None, callbacks: FlatlandCallbacks | None = None, tqdm_kwargs: dict | None = None, start_step: int = 0, end_step: int | None = None, fork_from_trajectory: Trajectory | None = None) → Trajectory

Creates trajectory by running submission (policy and obs builder).

Always backs up the actions and positions for steps executed in the tsvs. Can start from existing trajectory.

Parameters

policyPolicy

the submission’s policy

data_dirPath

the path to write the trajectory to

env: RailEnv

directly inject env, skip env generation

n_agents: int

number of agents

x_dim: int

number of columns

y_dim: int

number of rows

n_cities: int

Max number of cities to build. The generator tries to achieve this numbers given all the parameters. Goes into sparse_rail_generator.

max_rail_pairs_in_city: int

Number of parallel tracks in the city. This represents the number of tracks in the train stations. Goes into sparse_rail_generator.

grid_mode: bool

How to distribute the cities in the path, either equally in a grid or random. Goes into sparse_rail_generator.

max_rails_between_cities: int

Max number of rails connecting to a city. This is only the number of connection points at city boarder.

malfunction_duration_min: int

Minimal duration of malfunction. Goes into ParamMalfunctionGen.

malfunction_duration_max: int

Max duration of malfunction. Goes into ParamMalfunctionGen.

malfunction_interval: int

Inverse of rate of malfunction occurrence. Goes into ParamMalfunctionGen.

speed_ratios: Dict[float, float]

Speed ratios of all agents. They are probabilities of all different speeds and have to add up to 1. Goes into sparse_line_generator. Defaults to {1.0: 0.25, 0.5: 0.25, 0.33: 0.25, 0.25: 0.25}.

seed: int

Initiate random seed generators. Goes into reset.

obs_builder: Optional[ObservationBuilder]

Defaults to TreeObsForRailEnv(max_depth=3, predictor=ShortestPathPredictorForRailEnv(max_depth=50))

snapshot_intervalint

interval to write pkl snapshots

ep_id: str

episode ID to store data under. If not provided, generate one.

callbacks: FlatlandCallbacks

callbacks to run during trajectory creation

tqdm_kwargs: dict

additional kwargs for tqdm

start_stepint

start evaluation from intermediate step incl. (requires snapshot to be present); take actions from start_step and first step executed is start_step + 1. Defaults to 0 with first elapsed step 1.

end_stepint

stop evaluation at intermediate step excl. Capped by env’s max_episode_steps

fork_from_trajectoryTrajectory

copy data from this trajectory up to start step and run policy from there on

Returns

Trajectory