OakEmu-RL v0.1.0

Appearance

MetricDropAutoRollbackConfiguration ​

Configuration for automatic rollback triggered by a metric dropping by a threshold over a period.

Compares the metric value at the start of the lookback window with the value at the end. If the metric has dropped by more than drop_threshold, a rollback is triggered.

Source

Properties ​

effective_rollback_steps ​

python
effective_rollback_steps: int

Get the effective number of steps to roll back by.

Returns:

  • int – rollback_steps if set, otherwise period_steps.

Source

Values ​

drop_threshold ​

python
drop_threshold: float = Field(gt=0.0)

The amount by which the metric must drop to trigger a rollback.

Source

metric ​

python
metric: str

The metric tag prefix to monitor, for example 'Actor/Episode_Reward'.

Source

period_steps ​

python
period_steps: int = Field(gt=0)

The number of actor steps over which to measure the drop.

Source

rollback_steps ​

python
rollback_steps: int | None = Field(default=None, gt=0)

The number of steps to roll back by. Defaults to period_steps when None.

Source

type ​

python
type: Literal['metric_drop'] = 'metric_drop'

Source