HavanaDiscreteSession

Symbolica documentation for getting started, symbolic expressions, numerical evaluation, pattern matching, and APIs in Python and Rust.

`HavanaDiscreteSession`

class HavanaDiscreteSession

Caller-owned native sector importance sampling. Pilot pauses retain this object; persistent checkpoints require frozen production. No worker pool is created.

Attributes

Name	Description
`checkpoint_available`
`complete`	Whether this pilot epoch or production allocation has completed.
`sector_probabilities`	(Native sector id, current selection probability) in native problem order.
`settings`
`stage`

`checkpoint_available`

HavanaDiscreteSession.checkpoint_available: builtins.bool

`complete`

HavanaDiscreteSession.complete: builtins.bool

Whether this pilot epoch or production allocation has completed.

`sector_probabilities`

HavanaDiscreteSession.sector_probabilities: builtins.list[tuple[builtins.int, builtins.float]]

(Native sector id, current selection probability) in native problem order.

`settings`

HavanaDiscreteSession.settings: HavanaDiscreteSettings

`stage`

HavanaDiscreteSession.stage: builtins.str

Methods

Name	Description
`_repr_html_`
`adapt_pilot`	Adapt only a completed pilot and start another independent native pilot epoch.
`checkpoint`	Persist accepted frozen-production batches and native replay state
`freeze_production`	Adapt a completed pilot, freeze both grids, and discard all pilot estimates.
`live_observation`	Native point-statistics preview, separate from accepted complete-batch covariance.
`observation`
`snapshot`
`step`	Execute at most max_batches

`_repr_html_`

HavanaDiscreteSession._repr_html_(slf: HavanaDiscreteSession) -> builtins.str

`adapt_pilot`

HavanaDiscreteSession.adapt_pilot(
    *,
    discrete_learning_rate: builtins.float = 1.0,
    continuous_learning_rate: builtins.float = 1.0,
) -> IntegrationSnapshot

Adapt only a completed pilot and start another independent native pilot epoch.

`checkpoint`

HavanaDiscreteSession.checkpoint() -> bytes

Persist accepted frozen-production batches and native replay state. Pilot training is deliberately not serialized; retain the session to resume it.

`freeze_production`

HavanaDiscreteSession.freeze_production(
    *,
    points_per_batch: builtins.int,
    batches: builtins.int,
    discrete_learning_rate: builtins.float = 1.0,
    continuous_learning_rate: builtins.float = 1.0,
) -> IntegrationSnapshot

Adapt a completed pilot, freeze both grids, and discard all pilot estimates.

`live_observation`

HavanaDiscreteSession.live_observation() -> LiveObservation

Native point-statistics preview, separate from accepted complete-batch covariance.

`observation`

HavanaDiscreteSession.observation() -> IntegrationObservation

`snapshot`

HavanaDiscreteSession.snapshot() -> IntegrationSnapshot

`step`

HavanaDiscreteSession.step(
    max_batches: builtins.int = 1,
    *,
    evaluation_batch_size: builtins.int = 256,
    observer: typing.Optional[typing.Any] = None,
) -> IntegrationSnapshot

Execute at most max_batches. False from observer pauses after an accepted global batch. Interrupted batches are retried, without partial statistics.