Search Techniques

Coordinates: Tuple[float, ...]

Coordinates are in pyATF represented as a tuple of float s.

Cost: float

We currently use float as cost type.

class pyatf.search_techniques.search_technique.SearchTechnique

Searches over multi-dimensional coordinate space \((0,1]^D\).

initialize(dimensionality: int)

Initializes the search technique.

Parameters:

dimensionality – “D” of the coordinate space

finalize()

Finalizes the search technique.

get_next_coordinates() → Set[Coordinates]

Returns the next coordinates in \((0,1]^D\) for which the costs are requested.

Function get_next_coordinates() is called by pyATF before each call to report_costs(...).

Returns:

coordinates in \((0,1]^D\)

report_costs(costs: Dict[Coordinates, Cost])

Processes costs for coordinates requested via function get_next_coordinates().

Function report_costs(...) is called by pyATF after each call to get_next_coordinates().

Parameters:

costs – coordinates mapped to their costs

Index: int

Index is represented in pyATF as an int value.

class pyatf.search_techniques.search_technique_1d.SearchTechnique1D

Searches over one-dimensional index space \(\{ 0 , ... , |SP|-1 \}\), where \(|SP|\) is the search space size.

initialize(search_space_size: int)

Initializes the search technique.

Parameters:

search_space_size – the total number of configurations in the search space

finalize()

Finalizes the search technique.

get_next_indices() → Set[Index]

Returns the next indices in \(\{ 0 , ... , |SP|-1 \}\) for which the costs are requested.

Function get_next_indices() is called by pyATF before each call to report_costs(...).

Returns:

indices in \(\{ 0 , ... , |SP|-1 \}\)

report_costs(costs: Dict[Index, Cost])

Processes costs for indices requested via function get_next_indices().

Function report_costs(...) is called by pyATF after each call to get_next_indices().

Parameters:

costs – indices mapped to their costs