rtac.ac_functionalities.result_processing

In this module results of the tournaments are processed and computations for the next tournament ar performed.

Functions

processing_factory(scenario, logs)

Class factory to return the initialized class with data structures appropriate to the RTAC method scenario.ac.

Classes

`AbstractResultProcessing`(scenario, logs)	Abstract class with functions to process tournament results.
`Contender`()	Helper object for setting trueskill ranks.
`ResultProcessing`(scenario, logs)	Processes results of previous tournament.
`ResultProcessingCPPL`(scenario, logs)	Process results of previous tournament.
`ResultProcessingpp`(scenario, logs)	Process results of prvious tournament.

class rtac.ac_functionalities.result_processing.AbstractResultProcessing(scenario: Namespace, logs: RTACLogs)[source]

Bases: ABC

Abstract class with functions to process tournament results.

Parameters:

scenario (argparse.Namespace) – Namespace containing all settings for the RTAC.
logs (RTACLogs) – Object containing loggers and logging functions.

get_contender_dict() → dict[slice(<class 'str'>, <class 'rtac.ac_functionalities.rtac_data.Configuration'>, None)][source]

Returns contender_dict.

Returns:: Configuration selected to run in next tournament/on next problem instance. Dictionary with configuration id as key and Configuration object as value.
Return type:: dict[str, Configuration]

init_data() → dict[slice(<class 'str'>, <class 'rtac.ac_functionalities.rtac_data.Configuration'>, None)][source]

Initialize tournament result processing data according to ReACTR implementation.

Returns:: Randomly selected contenders.
Return type:: dict[str, Configuration]

abstract manage_pool() → None[source]

Replace contenders in pool if necessary.

Return type:: None

abstract process_results(rtac_data: RTACData | RTACDatapp, instance: str | None = None, tourn_nr: int | None = None) → None[source]

Perform tournament result processing necessary to replace contenders in pool and select contenders for next tournament/problem instance.

Parameters:

rtac_data (RTACData | RTACDatapp) – Object containing data and objects necessary throughout the rtac modules.
instance (str) – ID of the problem instance. Defaults to None.
tourn_nr (int) – Tournament number. Defaults to None.

Return type:

None

process_tourn(rtac_data: RTACData | RTACDatapp, instance: str | None = None, tourn_nr: int | None = None) → str | None[source]

Manage result processing.

rtac_data: RTACData | RTACDatapp: Object containing data and objects necessary throughout the RTAC modules.
instancestr: ID of the configuration to have won the previous tournament. Defaults to None.
tourn_nr: int: Number of the tournament that is processed.

Returns:: ID of the winner or None if problem instance could not be solved.
Return type:: str or None

result_summary_terminal(results: list[float], tourn_nr: int | None = None) → None[source]

Print the sum of the winner results of all tournaments so far.

Parameters:

results (list[float]) – List of results of previous tournament.
tourn_nr (int) – Number of previous tournament. Defaults to None.

Return type:

None

abstract select_contenders() → None[source]

Select contenders for next tournament/problem instance.

Return type:: None

class rtac.ac_functionalities.result_processing.Contender[source]

Bases: object

Helper object for setting trueskill ranks.

class rtac.ac_functionalities.result_processing.ResultProcessing(scenario: Namespace, logs: RTACLogs)[source]

Bases: AbstractResultProcessing

Processes results of previous tournament.

Note

Implementation based on the paper: “ReACTR: Realtime Algorithm Configuration through Tournament Rankings”

Parameters:

scenario (argparse.Namespace) – Namespace containing all settings for the RTAC.
logs (RTACLogs) – Object containing loggers and logging functions.

find_best(current_dict: list, number: int) → list[source]

Find the configurations scored as best by trueskill.

Parameters:

current_dict (list) – Dictionary with configuration ids as key and trueskill scores (mu, sigma) as value.
n_best (int) – Number of best configurations to draw.

Returns:

List of best configurations.

Return type:

list

get_winner(times: list[float], res: list[float]) → tuple[int, list[int]][source]

Get index of the winning configuration and ranks.

Parameters:

times (list[float]) – List of time results of previous tournament.
res (list[float]) – List of objective results of previous tournament.

Returns:

winner : int, Index of the winner.
ranks : list[int], Ranks of the contenders.

Return type:

tuple

init_scores() → None[source]

Initialize scores dict for trueskill.

Return type:: None

manage_pool() → None[source]

Replace contenders in pool according to Mu and Sigma (TrueSKill).

Return type:: None

process_results(rtac_data: RTACData | RTACDatapp, instance: str | None = None, tourn_nr: int | None = None) → None[source]

Perform tournament result processing necessary to replace contenders in the pool and select contenders for the next tournament/problem instance according to the ReACTR implementation.

Parameters:

rtac_data (RTACData | RTACDatapp) – Object containing data and objects necessary throughout the rtac modules.
instance (str) – ID of the configuration to have won the previous tournament. Defaults to None.
tourn_nr (int) – Number of the tournament that is processed.

Return type:

None

select_contenders() → None[source]

Select scenario.contenders == #contenders for next tournament/problem instance: top number of ‘self.scenario.keeptop’ and the rest randomly.

Return type:: None

class rtac.ac_functionalities.result_processing.ResultProcessingCPPL(scenario: Namespace, logs: RTACLogs)[source]

Bases: AbstractResultProcessing

Process results of previous tournament.

Parameters:

scenario (argparse.Namespace) – Namespace containing all settings for the RTAC.
logs (RTACLogs) – Object containing loggers and logging functions.

cppl_manage_pool(contender_dict: dict[str, rtac.ac_functionalities.rtac_data.Configuration], queue: Queue) → None[source]

Replace contenders in pool if necessary. Runs in separate thread.

Parameters:

contender_dict (dict[str, Configuration]) – List of contenders in the upcoming tournament. Loaded from logas or randomly selected.
queue (multiprocessing.Queue) – Queue to send data back to main thread.

Return type:

None

cppl_process_results(contender_dict: dict[str, rtac.ac_functionalities.rtac_data.Configuration], scenario: Namespace, rtac_data: RTACData | RTACDatapp, instance: str, pool: dict[str, rtac.ac_functionalities.rtac_data.Configuration], ob: bool, queue: Queue) → None[source]

Perform tournament result processing necessary to replace contenders in the pool and select contenders for the next tournament/problem instance within a separate thread.

Parameters:

contender_dict (dict[str, Configuration]) – List of contenders in the upcoming tournament. Loaded from logas or randomly selected.
scenario (argparse.Namespace) – Namespace containing all settings for the RTAC.
rtac_data (RTACData | RTACDatapp) – Object containing data and objects necessary throughout the rtac modules.
instance (str) – ID of the configuration to have won the previous tournament. Defaults to None.
pool (dict[str, Configuration]) – Initial pool of Configurations. Loaded from logas or randomly generated.
ob (bool) – True if scenario is objective value minimization, False if not.
queue (multiprocessing.Queue) – Queue to send data back to main thread.

Return type:

None

cppl_select_contenders(queue: Queue) → None[source]

Select contenders for next tournament/problem instance.

Parameters:: queue (multiprocessing.Queue) – Queue to send data back to main thread.
Return type:: None

init_cppl(scenario: Namespace, pool: dict[str, rtac.ac_functionalities.rtac_data.Configuration], random_config_gen: RandomConfigGen, contender_dict: dict[str, rtac.ac_functionalities.rtac_data.Configuration], queue: Queue) → None[source]

Initialize cppl in isolated mode. The bandit functions are called and executed in a separate thread so no build up in background threads of libraries that numpy, scipy and sklearn depend on is possible.

Parameters:

scenario (argparse.Namespace) – Namespace containing all settings for the RTAC.
pool (dict[str, Configuration]) – Initial pool of Configurations. Loaded from logas or randomly generated.
random_config_gen (RandomConfigGen) – Random configuration generator.
contender_dict (dict[str, Configuration]) – List of contenders in the upcoming tournament. Loaded from logas or randomly selected.
queue (multiprocessing.Queue) – Queue to send data back to main thread.

Return type:

None

manage_pool() → None[source]

Replace contenders in pool if necessary.

Return type:: None

process_results(rtac_data: RTACData | RTACDatapp, instance: str | None = None, tourn_nr: int | None = None) → None[source]

Perform tournament result processing necessary to replace contenders in the pool and select contenders for the next tournament/problem instance.

Parameters:

rtac_data (RTACData | RTACDatapp) – Object containing data and objects necessary throughout the rtac modules.
instance (str) – ID of the configuration to have won the previous tournament. Defaults to None.
tourn_nr (int) – Number of the tournament that is processed.

Return type:

None

process_tourn(rtac_data: RTACData | RTACDatapp, instance: str | None = None, tourn_nr: int | None = None) → str[source]

Manage result processing.

Parameters:

rtac_data (RTACData | RTACDatapp) – Object containing data and objects necessary throughout the rtac modules.
instance (str) – ID of the configuration to have won the previous tournament. Defaults to None.
tourn_nr (int) – Number of the tournament that is processed. Defaults to None.

Returns:

ID of the configuration to have won the previous tournament.

Return type:

str

select_contenders() → None[source]

Select contenders for next tournament/problem instance.

Return type:: None

class rtac.ac_functionalities.result_processing.ResultProcessingpp(scenario: Namespace, logs: RTACLogs)[source]

Bases: ResultProcessing

Process results of prvious tournament.

duplicates(ranks: list[int], rank: float) → list[int][source]

List the indices of the result in the results list.

Parameters:

ranks (list[int]) – List of objective results of previous tournament.
rank (float) – A single result.

Returns:

List of indices of this result in the results list.

Return type:

list[int]

get_winner(times: list[float], res: list[float]) → tuple[int, list[int]][source]

Get index of the winning configuration including last known intermediate outputs to break ties. Additionally outputs complete ranking to compute more detailed assessment with trueskill.

Parameters:

times (list[float]) – List of time results of previous tournament.
res (list[float]) – List of objective results of previous tournament.

Returns:

winner : int, Index of the winner.
ranks : list[int], Ranks of the contenders..

Return type:

tuple

rtac.ac_functionalities.result_processing.processing_factory(scenario: Namespace, logs: RTACLogs) → ResultProcessing | ResultProcessingpp | ResultProcessingCPPL[source]

Class factory to return the initialized class with data structures appropriate to the RTAC method scenario.ac.

Parameters:

scenario (argparse.Namespace) – Namespace containing all settings for the RTAC.
logs (RTACLogs) – Object containing loggers and logging functions.

Returns:

Initialized BaseTARunner object matching the RTAC method of the scenario.

Return type:

ResultProcessing or ResultProcessingpp or ResultProcessingCPPL