API

Documentation of main public modules.

class tengp.FunctionSet

Class containing functions which CGP system can use.

__getitem__(index)

Returns a function at given index.

__init__()

Initialize empty function set.

__len__()

Returns a number of functions in function set.

add(function, arity)

Add given function to function set.

Parameters:

• function (callable) – function to add (must have a workings __name__)
• arity (int) – arity (number of arguments) of given function

class tengp.Parameters(n_inputs, n_outputs, n_rows, n_columns, function_set, fitness_of_invalid=inf, use_tensors=False, max_back=None, cf_individual=False)

Contains construction parameters for CGP individual.

__init__(n_inputs, n_outputs, n_rows, n_columns, function_set, fitness_of_invalid=inf, use_tensors=False, max_back=None, cf_individual=False)

Creates Parameters object.

Parameters:

• n_inputs (int) – Number of input nodes (i.e. attributes)
• n_outputs (int) – Number of output nodes
• n_rows (int) – Number of rows
• n_columns (int) – Number of columns
• function_set (FunctionSet) – instance of FunctionSet class
• fitness_of_invalid (number) – penalty fitness assigned to individual producing error
• use_tensors (bool) – if true, CGP works with tensors instead of numpy arrays
• max_back (bool) – number of previous nodes (including inputs) to which node can connect. When set to None, not limit is imposed.
• cf_individual (bool) – if set to True, cost function recieves ground truth and an individual object as an argument. Otherwise it recieves ground truth and input transformed by individual.

Returns:

Parameters class instance.

tengp.simple_es(X, y, cost_function, params, target_fitness=None, population_size=5, evaluations=5000, random_state=None, mutation='point', mutation_probability=0.25, verbose=False, log=None, seed_individual=None)

Optimize a CGP system using a simple evolutionary strategy.

Parameters:

• X (numpy.ndarray) – input data, number of columns have to match the n_inputs parameter of Parameters object
• y (numpy.ndarray) – target output data, number of columns have to match the n_outputs parameter of Parameters object
• cost_function (callable) – cost function to minimize. It has two arguments (y_true, y_pred), where y_true is target output data and y_pred is output of CGP individual
• params (Parameters) – instance of Parameters class
• target_fitness (number or None) – fitness, at which evolution will stop. If None, it is not considered
• population_size (int) – size of population including parent
• evaluations (int) – maximum number of cost function evaluations
• random_state (int) – seed for random number generator
• mutation (string) – type of mutation to use, accept values ‘point’, ‘active’, ‘single’, ‘probabilistic’
• mutation_probability (float) – probability of mutating a given gene ( used only when mutation argument is set to ‘probabilistic’
• verbose (bool) – if True, outputs evolution info every 100 generations
• log (list) – if provided with a list, best fitness of each generation is stored here
• seed_individual (Individual) – if provided with instance of Individual class, the initial population is created according to this object - parent of first generation.

Returns:

List of individuals, the last generation of evolution

class tengp.individual.NPIndividual(genes, bounds, params)

transform(X)

Transforms the input data with expression encoded in individual.

Parameters:X (array-like) – 2D Numpy array, or tensor (if use_tensors was set to true in Parameters) Returns:Transformed data. If use_tensors was set to true, then list containing output tensors is returned. Otherwise Numpy array is returned.