Functions for experiments and utilities

Utilities

Preparing the datasets

pyridge.util.preprocess.prepare_data(folder, dataset, sep='\\s+', scaler='standard')

Read the data from the files and scale them. Target is supposed to be at the last column.

Parameters

• folder (str) – name of the folder where the dataset is.

• dataset (str) – name of the dataset to load.

• sep (str) – separator in string form. Default is spaces.

• scaler –

Returns

Activation and kernel functions

pyridge.util.activation.sigmoid(x)

Sigmoid function. It can be replaced with scipy.special.expit.

Parameters

x –

Returns

pyridge.util.activation.sigmoid_der(x)

Derivative of the sigmoid function.

Parameters

y –

Returns

pyridge.util.activation.linear_kernel(gamma: float = 1.0, X=None, Y=None)

Parameters

• gamma –

• X –

• Y –

Returns

pyridge.util.activation.rbf_kernel(gamma: float = 1.0, X=None, Y=None)

Function to obtain omega matrix.

Parameters

• gamma (float) –

• X (np.matrix) –

• Y –

Returns

pyridge.util.activation.u_dot_norm(u)

Return u vector with norm = 1.

Parameters

u –

Returns

Cross validation

pyridge.util.cross.cross_validation(predictor, train_data, train_target, hyperparameter, metric='accuracy', n_folds=5)

Cross validation training in order to find best parameter.

Parameters

• predictor –

• train_data –

• train_target –

• hyperparameter (dict) –

• metric (str) –

• n_folds (int) –

Returns

pyridge.util.cross.train_predictor(predictor, train_data, train_target, hyperparameter, metric='accuracy', n_folds=5)

Cross validation training in order to find best parameter.

Parameters

• predictor –

• train_data –

• train_target –

• hyperparameter (dict) –

• metric (str) –

• n_folds (int) –

Returns

Metrics

pyridge.util.metric.accuracy(clf, pred_data, real_targ)

Percentage of predicted targets that actually coincide with real targets.

Parameters

• clf – classifier with predict method.

• pred_data – array of the targets according to the classifier.

• real_targ (bool) – array of the real targets.

• real_targ – array of the real targets.

Returns

pyridge.util.metric.rmse(clf, pred_data, real_targ)

Parameters

• clf –

• pred_data –

• real_targ –

Returns

pyridge.util.metric.diversity(clf, pred_data=None, real_targ=None)

Implemented directly from MATLAB, not pythonic.

Parameters

• clf – Predictor.

• pred_data – Not used.

• real_targ – Not used.

Returns

Experiments

In order to perform several experiments and tests de predictors, generic test function is used for different algorithms and cross-validation hyperparameters.

pyridge.experiment.check.check_fold(folder_dataset='data/iris', train_dataset='train_iris.0', test_dataset='test_iris.0', sep='\\s+', algorithm='ELM', metric_list=['accuracy', 'rmse'], metric_cross='accuracy', hyperparameter=None, classification=True, autoencoder=False)

Generic test.

Parameters

• folder_dataset (str) –

• train_dataset (str) –

• test_dataset (str) –

• sep (str) –

• algorithm (str) –

• metric_list (list) –

• hyperparameter (dict) –

• metric_cross (str) –

• classification (bool) – True if we want a classification; False if we are looking for a regression.

• autoencoder (bool) – True if we want autoencoder test; False if we are looking for a classic supervised test.

Returns

a dictionary, with the metrics.

pyridge.experiment.check.check_algorithm(folder, dataset, algorithm, metric_list, hyperparameter, metric_cross=None, classification=True, autoencoder=False)

Testing easily a predictor along all the folds.

Parameters

• folder (str) –

• dataset (str) –

• algorithm (str) –

• metric_list (list) –

• metric_cross (str) –

• hyperparameter (dict) –

• classification (bool) – True if we want a classification; False if we are looking for a regression.

• autoencoder (bool) – True if we want autoencoder test; False if we are looking for a classic supervised test.

Returns

a dictionary, with the metrics.