dynex_qboost package

Submodules

dynex_qboost.qboost module

class dynex_qboost.qboost.AllStumpsClassifier(X, y)

Bases: dynex_qboost.qboost.EnsembleClassifier

Ensemble classifier with one decision stump for each feature.

class dynex_qboost.qboost.DecisionStumpClassifier(X, y, feature_index)

Bases: object

Decision tree classifier that operates on a single feature with a single splitting rule.

The index of the feature used in the decision rule is stored relative to the original data frame.

predict(X)

Predict class.

Parameters

X (array) – 2D array of feature vectors. Note that the array contains all features, while the weak classifier itself will make a prediction based only a single feature.

Returns

Array of class labels.

class dynex_qboost.qboost.EnsembleClassifier(weak_classifiers, weights, weak_classifier_scaling, offset=1e-09)

Bases: object

Ensemble of weak classifiers.

fit_offset(X)

Fit offset value based on class-balanced feature vectors.

Currently, this assumes that the feature vectors in X correspond to an even split between both classes.

get_selected_features()

Return list of features corresponding to the selected weak classifiers.

predict(X)

Compute ensemble prediction.

Note that this function returns the numerical value of the ensemble predictor, not the class label. The predicted class is sign(predict()).

predict_class(X)

Compute ensemble prediction of class label.

score(X, y)

Compute accuracy score on given data.

squared_error(X, y)

Compute squared error between predicted and true labels.

Provided for testing purposes.

class dynex_qboost.qboost.QBoostClassifier(X, y, lam, weak_clf_scale=None, drop_unused=True, num_reads=10000, annealing_time=300, mainnet=True)

Bases: dynex_qboost.qboost.EnsembleClassifier

Construct an ensemble classifier using quadratic loss minimization.

report_baseline(X, y)

Report accuracy of weak classifiers.

This provides context for interpreting the performance of the boosted classifier.

dynex_qboost.qboost.qboost_lambda_sweep(X, y, lambda_vals, val_fraction=0.4, verbose=False, **kwargs)

Run QBoost using a series of lambda values and check accuracy against a validation set.

Parameters

• X (array) – 2D array of feature vectors.

• y (array) – 1D array of class labels (+/- 1).

• lambda_vals (array) – Array of values for regularization parameter, lambda.

• val_fraction (float) – Fraction of given data to set aside for validation.

• verbose (bool) – Print out diagnostic information to screen.

• kwargs – Passed to QBoost.__init__.

Returns

QBoost instance with best validation score. lambda:

Lambda value corresponding to the best validation score.

Return type

QBoostClassifier

Module contents