panther.tuner.SkAutoTuner.Searching package

panther.tuner.SkAutoTuner.Searching package#

Submodules#

panther.tuner.SkAutoTuner.Searching.GridSearch module#

class panther.tuner.SkAutoTuner.Searching.GridSearch.GridSearch(max_iterations=10)[source]#

Bases: SearchAlgorithm

Grid search algorithm that systematically tries all combinations of parameters up to a maximum number of iterations.

This implementation performs a search through a pre-generated grid of parameter combinations. It iterates through these combinations and keeps track of the best parameters found and their corresponding score.

max_iterations#: The maximum number of parameter combinations to try.

param_space#: Dictionary mapping parameter names to lists of their possible values.

indexed_param_space#: A list of dictionaries, where each dictionary represents a unique combination of parameter values.

curr_iteration#: The current iteration number, indicating which parameter combination is being evaluated or will be evaluated next.

best_score#: The highest score achieved so far during the search.

best_params#: The dictionary of parameters that achieved the best_score.

__init__(max_iterations=10)[source]#

Initialize the GridSearch algorithm.

Parameters:: max_iterations – The maximum number of iterations to run.

property current_idx#: Alias for curr_iteration for backward compatibility.

property param_combinations#: Alias for indexed_param_space for backward compatibility.

initialize(param_space)[source]#

Initialize the search algorithm with the parameter space.

Parameters:: param_space – Dictionary of parameter names and their possible values

get_next_params()[source]#

Get the next set of parameters to try.

Returns:: Dictionary of parameter names and values to try, or None if finished

update(params, score)[source]#

Update the search algorithm with the results of the latest trial.

Parameters:

params – Dictionary of parameter names and values that were tried
score – The evaluation score for the parameters

save_state(filepath)[source]#

Save the current state of the search algorithm to a file.

Parameters:: filepath – The path to the file where the state should be saved.

load_state(filepath)[source]#

Load the state of the search algorithm from a file.

Parameters:: filepath – The path to the file from which the state should be loaded.

get_best_params()[source]#

Get the best set of parameters found so far.

Returns:: Dictionary of the best parameter names and values, or None if no params yet.

get_best_score()[source]#

Get the best score achieved so far.

Returns:: The best score, or None if no score yet.

reset()[source]#: Reset the search algorithm to its initial state while preserving param_space. If param_space is set, it regenerates the combinations.

is_finished()[source]#

Check if the search algorithm has finished its search (e.g., budget exhausted).

Returns:: True if the search is finished, False otherwise.

panther.tuner.SkAutoTuner.Searching.OptunaSearch module#

panther.tuner.SkAutoTuner.Searching.SearchAlgorithm module#

class panther.tuner.SkAutoTuner.Searching.SearchAlgorithm.SearchAlgorithm[source]#

Bases: ABC

Abstract base class for search algorithms to use in autotuning.

abstractmethod initialize(param_space)[source]#

Initialize the search algorithm with the parameter space.

Parameters:: param_space – Dictionary of parameter names and their possible values. Values can be lists (legacy) or ParamSpec types (Categorical, Int, Float).

abstractmethod get_next_params()[source]#

Get the next set of parameters to try.

Returns:: Dictionary of parameter names and values to try, or None if finished

abstractmethod update(params, score)[source]#

Update the search algorithm with the results of the latest trial.

Parameters:

params – Dictionary of parameter names and values that were tried
score – The evaluation score for the parameters

abstractmethod save_state(filepath)[source]#

Save the current state of the search algorithm to a file.

Parameters:: filepath – The path to the file where the state should be saved.

abstractmethod load_state(filepath)[source]#

Load the state of the search algorithm from a file.

Parameters:: filepath – The path to the file from which the state should be loaded.

abstractmethod get_best_params()[source]#

Get the best set of parameters found so far.

Returns:: Dictionary of the best parameter names and values, or None if no params yet.

abstractmethod get_best_score()[source]#