Basic usage

The file examples/basics.py showcases the major simulation-side components of AMaze:

• creating a maze, robot and simulation

• loading from an existing pre-trained agent

• having the agent explore said maze

• print various statistics about the maze and the agent’s performance

from amaze.simu import Maze, Robot, Simulation, load

We start by importing modules: everything related to the simulation is conveniently exposed by the module amaze.simu.

Note

pprint is imported to generate pretty outputs and is not mandatory.

maze = Maze.from_string("M16_10x10_U")
robot = Robot.BuildData.from_string("DD")
simulation = Simulation(maze, robot)

With every relevant class in local scope we can create the basic components: the Maze, Robot and Simulation. The first can be quickly generated from a string (see Maze.from_string()). For convenience, the inputs and outputs spaces can also be provided as a string (Robot.BuildData.from_string()). The last component is the simulation object, created using both the maze and robot description.

print("Maze stats:", pprint.pformat(maze.stats()))

For the curious, line 10 will provide detailed stats about the maze including its size, intersections, etc.

pprint.pprint(simulation.compute_metrics(maze, robot.inputs, robot.vision))

For the complexity metrics, which are theoretically dependant on the agent, one may use the simulation’s dedicated method to compute_metrics. Note that, contrary to the statistics, this implies a fair amount of computation (O(n*m), for a maze of n rows and m columns)

controller = load("examples/agents/unicursive_tabular.zip")
print(f"Agent infos: {pprint.pformat(controller.infos)}")

Next we retrieve one of the demonstration agents provided with the library and print some of the information stored alongside it.

simulation.run(controller)

After which we let the simulation run freely until completion. This means either success() (the agent has reached the target) or failure() (the agent has exceeded the deadline).

print("Target reached:", simulation.success())
print("Raw reward:", simulation.cumulative_reward())
print("Normalized reward:", simulation.normalized_reward())
print(f"Details: {pprint.pformat(simulation.infos())}")

To conclude, we extract select pieces of information from the simulation (success, rewards) as well as the more exhaustive infos(). The full listing for the example is shown below

Full listing for examples/basics.py

import pprint

from amaze.simu import Maze, Robot, Simulation, load

print("=" * 80)
maze = Maze.from_string("M16_10x10_U")
robot = Robot.BuildData.from_string("DD")
simulation = Simulation(maze, robot)

print("Maze stats:", pprint.pformat(maze.stats()))
print()
print("Maze complexity:")
pprint.pprint(simulation.compute_metrics(maze, robot.inputs, robot.vision))

print("=" * 80)
controller = load("examples/agents/unicursive_tabular.zip")
print(f"Agent infos: {pprint.pformat(controller.infos)}")

simulation.run(controller)

print("=" * 80)
print("Target reached:", simulation.success())
print("Raw reward:", simulation.cumulative_reward())
print("Normalized reward:", simulation.normalized_reward())
print(f"Details: {pprint.pformat(simulation.infos())}")