UGParameterEstimator.evaluators package

Submodules

UGParameterEstimator.evaluators.clusterEvaluator module

class UGParameterEstimator.evaluators.clusterEvaluator.ClusterEvaluator(luafilename, directory, parametermanager: ParameterManager, evaluation_type, parameter_output_adapter: ParameterOutputAdapter, fixedparameters={}, threadcount=10, cliparameters=[], ugsubmitparameters=[])

Bases: Evaluator

Evaluator for Clusters supporting UGSUBMIT.

Implements the Evaluator AbcstractBaseClass. Calls UGSUBMIT and UGINFO using the subprocess module to schedule tasks and get infos about them.

Implements a handler to catch unexpected program interruption while evaluating (for example, if the user cancels the operation). In this case, all open jobs will be cancelled.

Output of UG4 is redirected into a separate <id>_ug_output.txt file.

evaluate(evaluationlist, transform=True, tag='')

Evaluates the parameters given in evaluationlist using UG4, and the adapters set in the constructor.

Parameters

• evaluationlist (list of numpy arrays) – parametersets to evaluate

• transform (boolean, optional) – wether to transform the parameters with parametermanager set in this object, defaults to true

• tag (string) – tag-string attached to all produced evaluations for analysis purposes

Returns

list of parsed evaluation objects with the type given in the constructor, or ErroredEvaluation

Return type

list of Evaluation

property parallelism

Returns the parallelism of the evaluator

Returns

parallelism of the evaluator

Return type

int

UGParameterEstimator.evaluators.evaluator module

class UGParameterEstimator.evaluators.evaluator.Evaluator

Bases: ABC

Evaluator abstract base class

Defines the interface for evaluators. Implements a cache to avoid unnecessary evaluations.

classmethod ConstructEvaluator(luafile, directory, parametermanager: ParameterManager, evaluation_type: Evaluation, parameter_output_adapter: ParameterOutputAdapter, fixedparameters={}, threadcount=10, cliparameters=[])

Factory method to construct a suitable evaluator.

If UGSUBMIT can be detected on the system, a ClusterEvaluator will be used, if not, a LocalEvaluator.

Parameters

• luafilename (string) – path to the luafile to call for every evaluation

• directory (string) – directory to use for exchanging data with UG4

• parametermanager (ParameterManager) – ParameterManager to transform the parameters/get parameter information

• evaluation_type (type implementing Evaluation) – TYPE the evaluation shoould be parsed as.

• parameter_output_adapter (ParameterOutputAdapter) – output adapter to write the parameters

• fixedparameters (dictionary<string, string|number>, optional) – optional dictionary of fixed parameters to pass

• jobcount (int, optional) – optional maximum number of parallel jobs to submit in UGSUBMIT, defaults to 10

• cliparameters (list of strings, optional) – list of command line parameters to append to subprocess call. use separate entries for places that would normally require a space.

cache = {}

cached_evaluation_count = 0

checkCache(parameters)

Checks the internal evaluation cache and returns the stored result, if there is one, or None, if not.

Parameters

parameters (numpy array) – parameters to check

Returns

Evaluation, if in cache, or None

Return type

Evaluation

abstract evaluate(evaluationlist, transform=True, tag='')

Evaluates the parameters given in evaluationlist using UG4, and the adapters set in the constructor.

Parameters

• evaluationlist (list of numpy arrays) – parametersets to evaluate

• transform (boolean, optional) – wether to transform the parameters with parametermanager set in this object, defaults to true

• tag (string) – tag-string attached to all produced evaluations for analysis purposes

Returns

list of parsed evaluation objects with the type given in the constructor, or ErroredEvaluation

Return type

list of Evaluation

getStatistics()

returns the internal statistics as a string representation :return: string with statistics information :rtype: string

handleNewEvaluations(evaluations, tag)

Updates the internal evaluation cache and writes evaluations and new caching statistics to the set result object.

Parameters

• evaluations (list of Evaluation) – evaluations to handle

• tag (string) – tag to store the evaluations under in the result object

abstract property parallelism

Returns the parallelism of the evaluator

Returns

parallelism of the evaluator

Return type

int

reset()

resets the internal cache and statistics

resultobj = None

serial_evaluation_count = 0

setResultObject(res)

Sets the result object to write statistics to.

Parameters

res (Result) – resultobject to set

total_evaluation_count = 0

UGParameterEstimator.evaluators.localEvaluator module

class UGParameterEstimator.evaluators.localEvaluator.LocalEvaluator(luafile, directory, parametermanager: ParameterManager, evaluation_type: Evaluation, parameter_output_adapter: ParameterOutputAdapter, fixedparameters={}, threadcount=10, cliparameters=[])

Bases: Evaluator

Evaluator for usage on PCs without UGSUBMIT.

Implements the Evaluator AbcstractBaseClass. Can use MPI for local speedup, if threadcount > 1 is passed. Output of UG4 is redirected into a separate <id>_ug_output.txt file.

evaluate(evaluationlist, transform=True, tag='')

Evaluates the parameters given in evaluationlist using UG4, and the adapters set in the constructor.

Parameters

• evaluationlist (list of numpy arrays) – parametersets to evaluate

• transform (boolean, optional) – wether to transform the parameters with parametermanager set in this object, defaults to true

• tag (string) – tag-string attached to all produced evaluations for analysis purposes

Returns

list of parsed evaluation objects with the type given in the constructor, or ErroredEvaluation

Return type

list of Evaluation

property parallelism

Returns the parallelism of the evaluator. here it is one, as only one evaluation is handled in parallel.

Returns

parallelism of the evaluator

Return type

int

Module contents