cdlib.ensemble.random_search

random_search(graph: <Mock id='140410181329248'>, method: Callable[[<Mock id='140410181329440'>, dict], object], parameters: list, quality_score: Callable[[<Mock id='140410181329536'>, object], object], instances: int = 10, aggregate: Callable[[list], object] = <built-in function max>) → tuple

Returns the optimal partition of the specified graph w.r.t. the selected algorithm and quality score over a randomized sample of the input parameters.

Parameters:

• method – community discovery method (from nclib.community)
• graph – networkx/igraph object
• parameters – list of Parameter and BoolParameter objects
• quality_score – a fitness function to evaluate the obtained partition (from nclib.evaluation)
• instances – number of randomly selected parameters configurations
• aggregate – function to select the best fitness value. Possible values: min/max

Returns:

at each call the generator yields a tuple composed by: the optimal configuration for the given algorithm, input paramters and fitness function; the obtained communities; the fitness score

Example:

>>> import networkx as nx
>>> from cdlib import algorithms, ensemble
>>> g = nx.karate_club_graph()
>>> resolution = ensemble.Parameter(name="resolution", start=0.1, end=1, step=0.1)
>>> randomize = ensemble.BoolParameter(name="randomize")
>>> communities, scoring = ensemble.random_search(graph=g, method=algorithms.louvain,
>>>                                                       parameters=[resolution, randomize],
>>>                                                       quality_score=evaluation.erdos_renyi_modularity,
>>>                                                       instances=5, aggregate=max)
>>> print(communities, scoring)