Search

The BaseSearch class defines the interface for search strategies that generate candidate pools for evaluation. Search strategies can operate on predefined datasets (DatasetSearch), generate new candidates using models (GeneratorSearch), or apply mutations to existing sequences (SingleMutantSearch).

class alf_core.optimizer.search.BaseSearch

Bases: ABC

Base class for all types of search methods.

Search methods provide a way of searching the design space in order to propose new candidates not yet observed in the training dataset. This can be achieved in multiple ways depending on the search protocol, or availability of ground truth labels.

get_metrics(state)

Get metrics for the search function.

Parameters:

state (State) – Current task state.

Return type:

dict[str, float]

Returns:

Dictionary of metric names to values. Returns empty dict by default; subclasses should override.

class alf_core.optimizer.search.DatasetSearch

Bases: BaseSearch

Offline search method based on a dataset defining the search pool.

get_metrics(state)

Return recall and regret metrics for the dataset search method.

Both metrics compare the initial candidate pool against the candidates acquired during the loop. The acquired-only set is reconstructed from state.history so the initial labelled seed (which is not part of the candidate pool) does not enter the comparison.

Parameters:

state (State) – Current task state.

Return type:

dict[str, float]

Returns:

Dictionary containing recall and regret metrics comparing the initial candidate pool to the acquired candidates. The metrics are omitted until at least one candidate has been acquired.

class alf_core.optimizer.search.GeneratorSearch(model)

Bases: BaseSearch

Search method based on a model generating samples.

class alf_core.optimizer.search.ModelProtocolSearch(model, protocol)

Bases: BaseSearch

Search method based on a model and a protocol to define the search pool.

class alf_core.optimizer.search.ProtocolSearch(protocol)

Bases: BaseSearch

Search method based on a function/procedure to define the search pool.

class alf_core.optimizer.search.SearchProtocol

Bases: object

Search protocol is used to define the search process.