Unity-Technologies/ml-agents · warning · TrainerConfigWarning

Your final lesson definition contains completion_criteria fo

Error message

Your final lesson definition contains completion_criteria for {parameter_name}.It will be ignored.

What it means

In settings.py, a curriculum's _check_lesson_chain validates lesson progressions for a (redundant) completion_criteria on the last lesson; by definition the final lesson is never 'completed' into another lesson, so its completion_criteria is pointless. ML-Agents issues a TrainerConfigWarning and ignores it rather than erroring.

Source

Thrown at ml-agents/mlagents/trainers/settings.py:552

    parameter.
    """

    curriculum: List[Lesson]

    @staticmethod
    def _check_lesson_chain(lessons, parameter_name):
        """
        Ensures that when using curriculum, all non-terminal lessons have a valid
        CompletionCriteria, and that the terminal lesson does not contain a CompletionCriteria.
        """
        num_lessons = len(lessons)
        for index, lesson in enumerate(lessons):
            if index < num_lessons - 1 and lesson.completion_criteria is None:
                raise TrainerConfigError(
                    f"A non-terminal lesson does not have a completion_criteria for {parameter_name}."
                )
            if index == num_lessons - 1 and lesson.completion_criteria is not None:
                warnings.warn(
                    f"Your final lesson definition contains completion_criteria for {parameter_name}."
                    f"It will be ignored.",
                    TrainerConfigWarning,
                )

    @staticmethod
    def structure(d: Mapping, t: type) -> Dict[str, "EnvironmentParameterSettings"]:
        """
        Helper method to structure a Dict of EnvironmentParameterSettings class. Meant
        to be registered with cattr.register_structure_hook() and called with
        cattr.structure().
        """
        if not isinstance(d, Mapping):
            raise TrainerConfigError(
                f"Unsupported parameter environment parameter settings {d}."
            )
        d_final: Dict[str, EnvironmentParameterSettings] = {}
        for environment_parameter, environment_parameter_config in d.items():

View on GitHub (pinned to 3ecb446f75)

Solutions

Remove the completion_criteria block from the final lesson in the curriculum YAML.
If the criteria was meant to gate a real lesson, move it to the second-to-last lesson or add another lesson after it.
Re-run training; the warning is safe to ignore if you leave it, the criteria simply has no effect.

Example fix

// before
lesson: 2
  value: 3.0
  completion_criteria:
    measure: reward
    behavior: MyBehavior
// after
lesson: 2
  value: 3.0

Defensive patterns

Strategy: try-catch

Validate before calling

import yaml

def validate_curriculum(path):
    cfg = yaml.safe_load(open(path))
    for name, behavior in cfg.get("behaviors", {}).items():
        for param, curriculum in behavior.get("curriculum", {}).items():
            lessons = curriculum.get("lessons", [])
            if lessons and "completion_criteria" in lessons[-1]:
                print(f"warning: last lesson for {param} in {name} has completion_criteria; it will be ignored")

Type guard

def terminal_lesson_has_criteria(lessons: list) -> bool:
    return bool(lessons) and lessons[-1].get("completion_criteria") is not None

Try / catch

import warnings
from mlagents.trainers.settings import TrainerConfigWarning
with warnings.catch_warnings(record=True) as caught:
    warnings.simplefilter("always")
    # load curriculum / start training
    for w in caught:
        if issubclass(w.category, TrainerConfigWarning) and "final lesson" in str(w.message):
            print("Remove completion_criteria from the terminal lesson.")

Prevention

Keep completion_criteria only on non-terminal lessons.
Template lessons so the final lesson omits criteria by construction.
Treat TrainerConfigWarning as a config lint failure in CI (warnings-as-errors during config load).
Review curricula after copy-paste edits to lesson blocks.

When it happens

Trigger: Defining a curriculum where the terminal (last) lesson for a parameter has a completion_criteria block; training then emits this warning during curriculum setup.

Common situations: Copy-pasting lesson definitions so the last lesson inherits a completion_criteria; converting an old curriculum format; auto-generating lessons where all lessons share a criteria template.

Related errors

AI-assisted analysis of Unity-Technologies/ml-agents@3ecb446f75 (2026-09-02). Data as JSON: /api/errors/1eb33d3281ffb74b. Report an issue: GitHub.