Evaluators

Base class

class evaluators.helpers.eval.EvalBase

Bases: ABC

abstractmethod __init__()

Abstract class for learning algorithms in the Adaptive, Hybrid Feature Selection framework.

abstractmethod fit(**kwargs) → None

Abstract method for fitting the learning algorithm.

Parameters:

kwargs – Keyword arguments specific to individual evaluators.

abstractmethod predict(X: ndarray) → ndarray

Abstract function for predicting the target vector on the given data.

Parameters:

X (np.ndarray) – Data to predict the target vector from.

Returns:

The predicted label vector.

Return type:

np.ndarray

score(X: ndarray, y: ndarray) → Metrics

Performs a prediction, then compares with the true class labels. Returns a Metrics object.

Parameters:

  • X (np.ndarray) – Data to predict the target vector from.

  • y (np.ndarray) – Ground truth target vector.

Returns:

Metrics object containing the predicted and true class labels.

Return type:

Metrics

ANN with Levenberg-Marquardt optimization

class evaluators.e_AgiLM.AgiLM(layers: list[[<class 'int'>, collections.abc.Callable[[float], float] | None]], tau: float = 1.0, weights_boundary: list[float] = [-0.1, 0.1])

Bases: EvalBase

__init__(layers: list[[<class 'int'>, collections.abc.Callable[[float], float] | None]], tau: float = 1.0, weights_boundary: list[float] = [-0.1, 0.1])

Implements the interface of an artificial neural network with Levenberg-Marquardt optimization made by VIHAROS, Zsolt János and SZŰCS, Ágnes. https://doi.org/10.48550/arXiv.2211.11491

Further improvements to this implementation were done by HOANG, Anh Tuan; VINCZE, Tibor; GERCUJ, Bence.

Parameters:

  • layers (list[[int, Callable[[float], float] | None], ]) – A two-dimensional list describing the network architecture. Each sublist defines neuron count and activation function. Example: [[8, None], [4, sigmoid], [2, sigmoid]] describes an ANN with 8 inputs, one hidden layer with 4 neurons and sigmoid activation, and binary output with sigmoid activation. “sigmoid” is a Python function with signature (float) -> float.

  • tau (float) – Tau parameter controlling the optimizer. Default value is 1.0.

  • weights_boundary (list[float, float]) – Boundary for weights. Default value is [-0.1, 0.1].

Returns:

None

Return type:

None

fit(**kwargs) → None

Fits the neural network on the provided data.

Parameters:

  • kwargs – See below for non-optional and relevant parameters. Consult the source code for the remaining arguments.

  • train_inputs – Data for the network to be fitted on. np.ndarray

  • train_targets – Target variable for training. np.ndarray

  • val_inputs – Data to test per epoch. np.ndarray

  • val_targets – Target variable for validation. np.ndarray

  • mu – Optimizer parameter. float

  • initial_tau – Optimizer parameter. float

  • max_iteration – Maximum number of epochs for training. int

  • early_max_stepsize – Early stopping patience. int

  • MSE_training – Whether to use Mean-Squared Error as a loss function. bool

  • fix – Whether to use the fixed version of the algorithm. bool

Returns:

None

Return type:

None

predict(X: ndarray) → ndarray

Performs prediction based on the input data. Fitting is needed prior to prediction.

Parameters:

X (np.ndarray) – Data from which the target variable is predicted from.

Returns:

Predicted target variable.

Return type:

np.ndarray

evaluators.e_AgiLM.sigmoid(x)