add-passive-algorithm

name	add-passive-algorithm
description	Add a new passive learning algorithm to AMLGym. Use when the user wants to integrate an algorithm that learns from pre-collected trajectory files.

Add a Passive Algorithm

Create a new passive algorithm adapter in amlgym/algorithms/.

Steps

1. Create amlgym/algorithms/$ARGUMENTS.py (class name must match filename, PascalCase, case-insensitive)

2. Inherit from PassiveAlgorithmAdapter and implement learn():

from dataclasses import dataclass
from typing import List
from amlgym.algorithms.PassiveAlgorithmAdapter import PassiveAlgorithmAdapter

@dataclass
class $ARGUMENTS(PassiveAlgorithmAdapter):
    """Docstring with paper reference and usage example."""

    # Optional: add paper reference metadata (used by registry for display)
    # _reference: ClassVar[OrderedDict[str, str]] = {
    #     'Authors': "Author names",
    #     'Title': "Paper title",
    #     'Venue': "Conference/journal",
    #     'Year': "2024",
    #     'URL': "https://...",
    # }

    # Algorithm parameters as dataclass fields with defaults
    my_param: float = 1.0

    def learn(self,
              domain_path: str,
              trajectory_paths: List[str]) -> str:
        # domain_path: input PDDL domain with predicates/operator signatures (no preconditions/effects)
        # trajectory_paths: list of trajectory file paths
        # Return: learned PDDL domain string
        ...

3. Add external dependencies to requirements.txt if needed

4. No registration needed — amlgym/algorithms/__init__.py auto-discovers all algorithm files

Reference implementation

Study amlgym/algorithms/SAM.py for the canonical passive adapter pattern:

• Preprocessing trajectories into algorithm-specific format
• Invoking the learning algorithm
• Cleaning up temp files
• Returning the learned PDDL model as a string

Trajectory format

Trajectory files use PDDL-like format with alternating (:state ...) and (:action ...) blocks. See amlgym/benchmarks/trajectories/ for examples.

Testing

from amlgym.algorithms import get_algorithm
from amlgym.benchmarks import get_domain_path, get_trajectories_path
from amlgym.util.util import empty_domain
from amlgym.metrics import syntactic_precision, syntactic_recall

domain = 'blocksworld'
agent = get_algorithm('$ARGUMENTS')
domain_path = get_domain_path(domain)
input_domain = empty_domain(domain_path)
traj_paths = get_trajectories_path(domain)
model = agent.learn(input_domain, traj_paths)

# Write learned model and evaluate
with open('learned.pddl', 'w') as f:
    f.write(model)
print("Precision:", syntactic_precision('learned.pddl', domain_path))
print("Recall:", syntactic_recall('learned.pddl', domain_path))