Active Learning Proposal Context — ALContext • celecx

Ephemeral context passed from OptimizerAL to ALProposer objects for a single outer proposal round.

The context keeps shallow references to the optimizer's canonical surrogate and acquisition-function registries. Access should go through $get_surrogate() and $get_acq(), because these methods implement lazy model updates, lazy acquisition-function fitting, and proposer-local pending-point filtering.

Creates a new active-learning context.

Adds points selected for the current proposal batch.

Lazily updates and returns a canonical surrogate.

Lazily wires, fits, updates, and returns an acquisition function.

Creates a fresh acquisition-function working object.

Points that have been evaluated or are pending evaluation.

Arguments

instance: (bbotk::EvalInstance)
Search or optimization instance.
surrogates: (named list() of mlr3mbo::Surrogate)
Canonical run-local surrogate objects.
acq_functions: (named list() of mlr3mbo::AcqFunction)
Unwired acquisition-function prototypes.
run_state: (list())
Run-local mutable state shared across proposal rounds.
allow_repeat_evaluations: (logical(1))
Whether to allow repeat evaluations of the same point.
xdt: (data.table)
Points selected in the current proposal round.
id: (character(1))
Surrogate registry id.
surrogate_id: (character(1))
Surrogate registry id.
clone: (logical(1))
Whether to return a fresh working clone.
acq_id: (character(1))
Acquisition-function registry id.
surrogate: (mlr3mbo::Surrogate)
Surrogate to wire into the acquisition function.
fit_scope: (character(1))
Either "global", "candidate", or "search_space". "global" falls back to "search_space" when the run has no finite pool.
pool: (NULL | data.table)
Candidate pool for candidate-scope fitting.

Value

mlr3mbo::Surrogate.

mlr3mbo::AcqFunction.

Fields

instance

(bbotk::EvalInstance)
Search or optimization instance.

pool

(NULL | data.table) Full finite candidate pool.

unevaluated_xdt

(NULL | data.table) Remaining finite candidate pool that were not evaluated yet (but may have been proposed already; use proposable_xdt to get candidates that can still be proposed). NULL for continuous contexts without pool.

'unevaluated' candidates are the union of 'proposable' and 'pending' candidates.

unevaluated_indices

(NULL | integer) Indices within pool of the remaining finite candidate pool that were not evaluated yet (but may have been proposed already; use proposable_indices to get indices of candidates that can still be proposed). NULL for continuous contexts without pool.

'unevaluated' indices are the union of 'proposable' and 'pending' indices.

evaluated_xdt

(data.table) Points that have been evaluated.

evaluated_indices

(integer | NULL) Indices within pool of points that have been evaluated. NULL for continuous contexts without pool.

pending_xdt

(data.table) Points already selected during the current proposal round. These are a subset of 'unevaluated' points.

pending_indices

(integer | NULL) Indices within pool of points already selected during the current proposal round. NULL for continuous contexts without pool. These are a subset of 'unevaluated' indices.

proposable_xdt

(NULL | data.table) Candidates that can still be proposed. NULL for continuous contexts without pool. These are all 'unevaluated' points that are not 'pending' when allow_repeat_evaluations is FALSE; otherwise, these are all points that are not pending.

proposable_indices

(integer | NULL) Indices within pool of candidates that can still be proposed. NULL for continuous contexts without pool. These are all 'unevaluated' indices that are not 'pending' when allow_repeat_evaluations is FALSE; otherwise, these are all indices that are not pending.

evaluated_and_pending_indices

(integer | NULL) Indices within pool of points that have been evaluated or are pending evaluation. NULL for continuous contexts without pool.

allow_repeat_evaluations

(logical(1)) Whether to allow repeat evaluations of the same point.