Unity-Technologies/ml-agents · error · UnityTrainerException

The schedule {self.schedule} is invalid.

Error message

The schedule {self.schedule} is invalid.

What it means

LearningRateSchedule.get_value raises UnityTrainerException when the configured schedule name is neither CONSTANT nor LINEAR. ML-Agents validates schedule types via the ScheduleType enum; an unrecognized value reaching get_value means an invalid hyperparameter got past (or bypassed) config validation.

Source

Thrown at ml-agents/mlagents/trainers/torch_entities/utils.py:99

            self.schedule = schedule
            self.initial_value = initial_value
            self.min_value = min_value
            self.max_step = max_step

        def get_value(self, global_step: int) -> float:
            """
            Get the value at a given global step.
            :param global_step: Step count.
            :returns: Decayed value at this global step.
            """
            if self.schedule == ScheduleType.CONSTANT:
                return self.initial_value
            elif self.schedule == ScheduleType.LINEAR:
                return ModelUtils.polynomial_decay(
                    self.initial_value, self.min_value, self.max_step, global_step
                )
            else:
                raise UnityTrainerException(f"The schedule {self.schedule} is invalid.")

    @staticmethod
    def polynomial_decay(
        initial_value: float,
        min_value: float,
        max_step: int,
        global_step: int,
        power: float = 1.0,
    ) -> float:
        """
        Get a decayed value based on a polynomial schedule, with respect to the current global step.
        :param initial_value: Initial value before decay.
        :param min_value: Decay value to this value by max_step.
        :param max_step: The final step count where the return value should equal min_value.
        :param global_step: The current step count.
        :param power: Power of polynomial decay. 1.0 (default) is a linear decay.
        :return: The current decayed value.
        """

View on GitHub (pinned to 3ecb446f75)

Solutions

1. Set learning_rate_schedule to exactly 'constant' or 'linear' in the trainer config YAML
2. Validate the config against the current ML-Agents trainer_config.schema.json before training
3. Upgrade/downgrade configs when migrating between ML-Agents versions (schedule options changed across releases)
4. If constructing programmatically, use ScheduleType.CONSTANT / ScheduleType.LINEAR enums instead of raw strings

Example fix

// before (trainer_config.yaml)
learning_rate_schedule: exponential
// after
learning_rate_schedule: linear

from mlagents.trainers.settings import ScheduleType
schedule = hyperparams.learning_rate_schedule
assert schedule in (ScheduleType.CONSTANT, ScheduleType.LINEAR), f"Invalid schedule {schedule}"

def is_valid_schedule(value) -> bool:
    try:
        return ScheduleType(value.lower()) in (ScheduleType.CONSTANT, ScheduleType.LINEAR)
    except (ValueError, AttributeError):
        return False

from mlagents.trainers.exception import UnityTrainerException
try:
    lr = lr_schedule.get_value(step)
except UnityTrainerException as e:
    logger.error(f"Bad schedule config: {e}")
    lr = hyperparams.learning_rate  # fall back to constant

When it happens

Trigger: Setting hyperparameters.learning_schedule / learning_rate_schedule to a string other than 'constant' or 'linear' in the trainer YAML config and constructing the schedule.

Common situations: Typo in the YAML (e.g. 'exponential' or 'Cosine'); editing a config copied from an older ML-Agents version whose schedule names changed; generating configs programmatically with raw strings instead of ScheduleType enum values.

Related errors

• The trainer config contains an unknown trainer type {trainer
• The trainer was unable to process any of the provided inputs
• The one of the goals uses variable length observations. This
• Trainer was unable to process any of the goals provided as i
• Visual observation resolution ({width}x{height}) is too smal

AI-assisted analysis of Unity-Technologies/ml-agents@3ecb446f75 (2026-09-02). Data as JSON: /api/errors/3549aa0fba53194f. Report an issue: GitHub.