machetli.pddl — PDDL Files

This package allows transforming PDDL input files. Usually, in a Machetli script you will use it in the following way:

from machetli import pddl
initial_state = pddl.generate_initial_state("path/to/domain.pddl", "path/to/problem.pddl")
successor_generators = [pddl.RemoveActions(), ...]

You can then start your Machetli search using the initial PDDL problem initial_state and a set of PDDL successor generators successor_generators. Finally, write out your results using

pddl.write_files(result, "path/to/result-domain.pddl", "path/to/result-problem.pddl")

where result is the value returned by the search function.

The successor generators described below denote possible transformations.

class machetli.pddl.RemoveActions

For each action schema in the PDDL domain, generate a successor where this action schema is removed. The order of the successors is randomized.

get_description()

get_successors(state)

Yield successors of state.

class machetli.pddl.RemoveObjects

For each object in the PDDL problem, generate a successor that removes this object from the PDDL task. The order of the successors is randomized.

get_description()

get_successors(state)

Yield successors of state.

class machetli.pddl.RemovePredicates(replace_with='dynamic')

For each predicate in the PDDL domain, generate a successor where this predicate is compiled away. This is accomplished by scanning the entire task for the atom to be removed, instantiating each instance of this atom with a constant according to replace_with:

• "true" replaces all atoms of the removed predicate with true,

• "false" replaces all atoms of the removed predicate with false, and

• "dynamic" (default) replaces an atom of the removed predicate with true if it occurs positively and with false otherwise.

The order of the successors is randomized.

get_description()

get_successors(state)

Yield successors of state.

machetli.pddl.generate_initial_state(domain_path: Union[pathlib.Path, str], task_path: Union[pathlib.Path, str]) → dict

Parse the PDDL task defined in the given PDDL files.

Returns

a dictionary pointing to the task specified in the files.

machetli.pddl.run_evaluator(evaluate)

Load the state passed to the script via its command line arguments, then run the given function evaluate on the domain and problem encoded in the state, and exit the program with the appropriate exit code. If the function returns True, use EXIT_CODE_BEHAVIOR_PRESENT, otherwise use EXIT_CODE_NOT_BEHAVIOR_PRESENT. In addition to running the evaluator, this function creates the PDDL files as ‘domain.pddl’ and ‘problem.pddl’ in the current directory.

This function is meant to be used as the main function of an evaluator script. Instead of a path to the state, the command line arguments can also be paths to a PDDL domain and problem (where the domain can be omitted if it can be found with automated naming rules). This is meant for testing and debugging the evaluator directly on PDDL input.

Parameters

evaluate – is a function taking filenames of a PDDL domain and problem file as input and returning True if the specified behavior occurs for the given instance, and False if it doesn’t. Other ways of exiting the function (exceptions, sys.exit with exit codes other than EXIT_CODE_BEHAVIOR_PRESENT or EXIT_CODE_BEHAVIOR_NOT_PRESENT) are treated as failed evaluations by the search.

machetli.pddl.write_files(state: dict, domain_path: Union[pathlib.Path, str], problem_path: Union[pathlib.Path, str])

Write the domain and problem files represented in state to disk.