Configuration

Contains the classes and functions related to abrain’s configuration

class abrain.Config

Wrapper for the C++ configuration data

Allows transparent access to the values used for CPPN structure, genome mutation (Genome.mutate()), and ANN/ES-HyperNEAT parameters

classmethod from_json(j: Dict)

Restore values from a json-compliant Python dictionary

Parameters:

j – the dictionary to parse values from

static known_function(name: str) → bool

Whether the requested function name is a built-in

classmethod read(path: Path)

Try to load data from provided path

Parameters:

path – Filename

classmethod show()

Write the configuration on standard output

static test_valid() → bool

Function used after reading a configuration file to ensure validity

classmethod to_json() → Dict

Convert to a json-compliant Python dictionary

classmethod write(path: Path | None)

Write the configuration to the specified file or stdout

Parameters:

path – where to write or none to print to screen

Bias = <ESHNOutputs.Bias: 2>

LEO = <ESHNOutputs.LEO: 1>

Weight = <ESHNOutputs.Weight: 0>

activationFunc = 'ssgn'

The activation function used by all hidden/output neurons (inputs are passthrough)

allowPerceptrons = True

Attempt to generate a perceptron if no hidden neurons were discovered

annWeightsRange = 3.0

Scaling factor s for the CPPN w output mapping \([-1,1] to [-s,s]\)

bndThr = 0.15000000596046448

Minimal divergence threshold for discovering neurons

cppnWeightBounds = Bounds(-3, -1, 1, 3, 0.01)

Initial and maximal bounds for each of the CPPN’s weights

defaultOutputFunction = 'id'

Output function for random generic CPPNs

divThr = 0.30000001192092896

Division threshold for a quad-/octtree cell/cube

eshnOutputFunctions = OutputFunctions{0: bsgm, 1: step, 2: id}

Functions used for CPPN2D and CPPN3D outputs

functionSet = Strings[abs, gaus, id, bsgm, sin, step]

List of functions accessible to nodes via creation/mutation

initialDepth = 2

Initial division depth for the underlying quad-/octtree

iterations = 10

Maximal number of discovery steps for Hidden/Hidden connections. Can stop early in case of convergence (no new neurons discovered)

maxDepth = 3

Maximal division depth for the underlying quad-/octtree

mutationRates = MutationRates{add_l: 0.0681818, add_n: 0.0454545, del_l: 0.0909091, del_n: 0.0681818, mut_f: 0.227273, mut_w: 0.5}

Probabilities for each point mutation (addition/deletion/alteration)

Glossary:

• add_l: add a random link between two nodes (feedforward only)

• add_n: replace a link by creating a node

• del_l: delete a random link (never leaves unconnected nodes)

• del_n: replace a simple node by a direct link

• mut_f: change the function of a node

• mut_w: change the connection weight of a link

varThr = 0.30000001192092896

Variance threshold for exploring a quad-/octtree cell/cube

class abrain.Config.ESHNOutputs

Members:

Weight

LEO

Bias

Bias = <ESHNOutputs.Bias: 2>

LEO = <ESHNOutputs.LEO: 1>

Weight = <ESHNOutputs.Weight: 0>

name

Name of the output

value

Corresponding value of the output

class Config.Strings(values: list[str])

A collections of strings (interfacing with the C++ layer)

class Config.MutationRates(values: dict[str, float])

A map from strings to floats (interfacing with the C++ layer)

class Config.OutputFunctions(values: dict[str, str])

A map from strings to function names (strings, interfacing with the C++ layer)

class Config.FBounds(min: float, rndMin: float, rndMax: float, max: float, stddev: float)

A wrapper for mutation bounds. Absolute range is [min, max]. Values produced through random initialization are further restricted to [rndMin, rndMax] with

\[min \leq rndMin \leq rndMax \leq max\]

stddev is the standard deviation for every point-mutation applied to the corresponding field.