Random Search on Pool-Restricted or Discrete Objectives

Optimizer that randomly samples configurations from a candidate pool when the objective has the "pool_restricted" property (typically by inheriting from ObjectivePoolAbstract), or from the full grid of a completely discrete search space (where all parameters are paradox::p_int, paradox::p_fct, or paradox::p_lgl).

When the search space is continuous (not fully discrete) and the objective is not pool-restricted, this optimizer falls back to uniform random search (like bbotk::OptimizerBatchRandomSearch).

Creates a new instance of this R6 class.

Arguments

grid_expansion_limit

(integer(1))
Upper limit on the number of rows to materialize for the full grid for discrete search spaces in Objectives that are not pool-restricted. An error is raised if the full grid would exceed this limit. This is to protect against accidental usage of too much memory. Default is 1e7L.

Parameters

batch_size

integer(1)
Number of configurations per batch. Inf evaluates the entire remaining pool in a single batch (forbidden when replace_samples = "within_batches" or when the objective is not pool-restricted). Default is 1.

max_batches

integer(1)
Maximum number of batches to evaluate. Inf means no batch limit (termination is governed solely by the instance's terminator). Default is Inf.

replace_samples

character(1)
Controls duplicate handling across evaluations:

• "never": configurations already present in the archive are excluded from sampling. The pool shrinks as evaluations accumulate.

• "between_batches": each batch is drawn without replacement from the full pool, but previous batches do not affect future sampling.

• "within_batches": samples inside each batch are drawn with replacement from the full pool, so duplicates are possible even within a single batch.

Default is "never".