sequince.steps.mutation.adaptive.policies.acceptance#
Choose mutation budgets from observed acceptance probabilities.
Classes#
Target expected accepted moves per particle using completed counts. |
Module Contents#
- class sequince.steps.mutation.adaptive.policies.acceptance.AcceptanceRatePolicy(target_steps, *, min_steps, max_steps, min_acceptance=1e-06)[source]#
Target expected accepted moves per particle using completed counts.
Choose
ceil(target_steps / max(acceptance, min_acceptance)), clipped to configured bounds. With zero attempts, retain the current budget or recover to one, subject to bounds. Fractional accepted counts are supported. Both APIs divide in floating point before the ceiling and requiremax_steps <= 2**24for safe integer conversion.- min_steps#
- max_steps#
- target_steps#
- min_acceptance#
- validate_functional_configuration()[source]#
Require bounds exactly representable in float32 and int32.
- Return type:
None
- initialise_functional(*, initial_steps, context)[source]#
- Parameters:
initial_steps (sequince.typing.Array)
context (sequince.steps.context.StepContext)
- Return type:
sequince.typing.Array
- update_functional(*, current_steps, metrics, context)[source]#
Choose a bounded budget from scalar completed-block counts.
Counts must be finite with
0 <= accepted <= attempted. Compiled code checks shapes only; callers must validate values. Eager calculations use host float64; other backend precisions may round differently.- Parameters:
current_steps (sequince.typing.Array)
metrics (sequince.steps.mutation.metrics.MutationMetricsOutput)
context (sequince.steps.context.StepContext)
- Return type:
sequince.typing.Array
- initialise(*, initial_steps, context)[source]#
- Parameters:
initial_steps (int)
context (sequince.steps.context.StepContext)
- Return type:
- update(*, current_steps, metrics, context)[source]#
- Parameters:
current_steps (int)
metrics (sequince.steps.mutation.metrics.MutationMetricsOutput)
context (sequince.steps.context.StepContext)
- Return type: