dataeval.performance.schedules.ManualSchedule¶

class dataeval.performance.schedules.ManualSchedule(eval_points)¶

Custom evaluation schedule with user-specified points.

Allows explicit control over evaluation points, useful for comparing specific dataset sizes or focusing on particular data regimes.

Parameters:¶

eval_points : int | Iterable[int] | NDArray[integer]¶

Evaluation points. Can be: - Single int: evaluate at one size - Iterable of ints: evaluate at multiple sizes - NumPy array: evaluate at array values

Raises:¶

ValueError – If eval_points contains non-numeric values

Examples

Single point:

>>> schedule = ManualSchedule(50)
>>> schedule.get_steps(dataset_length=100)
array([50])

Multiple points:

>>> schedule = ManualSchedule([10, 50, 100])
>>> schedule.get_steps(dataset_length=100)
array([ 10,  50, 100])

get_steps(dataset_length)¶

Return the pre-specified evaluation points.