Selecting and Looping over TestFunctions

Running an experiment boils down to running an implemented procedure over multiple TestFunctions. This can be rather boring to code, so a container class functions.FunctionFeeder has been implemented that allows you to select groups of testunctions and to loop over these functions for your experiment.

Selecting individual testfunctions

The FunctionFeeder container class is part of the functions module:

import high_dimensional_sampling as hds
feeder = hds.functions.FunctionFeeder()

The FunctionFeeder can load individual functions by function name. For example

feeder.load_function('Cosine')

This adds the Cosine test function to the feeder container. Some testfunctions have configuration parameters, which can be set through an additional input argument to the load_function method. For example:

feeder.load_function('Rosenbrock', {'dimensionality': 3})

Selecting multiple testfunctions at the same time

All implemented testfunctions are part of function groups. You can select all functions in a group by loading the group. To load all functions with a implemented derivative for example, you can run

feeder.load_function_group('with_derivative')

It is also possible to select intersections of multiple groups. For instance: if you want to select all functions with an implemented derivative that are meant for optimisation problems, the following code can be used:

feeder.load_function_group(['with_derivative','optimisation'])

A second argument can be added to this function to add configuration to the loaded testfunctions.

config = {
    'Rosenbrock': {'dimensionality': 3},
    'Rastrigin': {'dimensionality': 6}
}
feeder.load_function_group('optimisation', config)

Implemented groups

In the table below you can find an overview of all implemented groups and which functions are part of these groups.

Function	optimisation	posterior	with_derivative	no_derivative	bounded	unbounded
Rastrigin	x		x		x
Rosenbrock	x			x		x
Beale	x			x	x
Booth	x			x	x
BukinNmbr6	x			x	x
Matyas	x			x	x
LeviNmbr13	x			x	x
Himmelblau	x			x	x
ThreeHumpCamel	x			x	x
Sphere	x		x			x
Ackley	x			x	x
Easom	x			x	x
Cosine		x	x		x
Block		x		x		x
Bessel		x	x		x
ModifiedBessel		x	x		x
Eggbox		x		x	x
MultivariateNormal		x		x	x
GaussianShells		x		x	x
Linear	x	x		x	x
BreitWigner		x	x		x
Reciprocal		x	x		x
Schwefel	x			x	x
GoldsteinPrice	x			x	x

Adding testfunctions by hand

Although the FunctionFeeder takes care of the correct initialisation and configuration of testfunctions, if you feel more comfortable with adding functions that you initialised yourself, you can do this as well. In that case, use the add_function method:

f = hds.functions.Cosine()
feeder.add_function(f)

Looping over testfunctions

After having selected and added all functions you want to loop over, you can simply loop over these functions with a for-loop. If you want to print all the function names, you can use the following code for example:

for function in feeder:
    print(type(function).__name__)