hdlib v0.1.18

hdlib v0.1.18

Add:

• The space dictionary as part of a Space object is now an OrderedDict, making Space objects iterable over their set of vectors;
• Python examples under the examples folder are now available as part of the hdlib package.

Fix:

• Fix dumping and loading Vector and Space objects to and from pickle files;
• space.Space.bulk_insert function now checks whether the names of the input vectors are instances of bool, int, float, str, and None before creating and inserting vectors into the space;
• Distance thresholds in space.Space.find and space.Space.find_all are now set to numpy.Inf by default.

hdlib v0.1.17

hdlib v0.1.17

Add:

• Add the subtraction operator to the arithmetic module;
• Add __sub__ to space.Vector that makes use of arithmetic.subtraction to element-wise subtract two Vector objects;
• Add space.Vector.subtraction to element-wise subtract a vector from a Vector object inplace;
• Add graph.Graph to build vector-symbolic representations of directed and undirected, weighted and unweighted graphs;
• Extend test/test.py with new unit tests.

hdlib v0.1.16

hdlib v0.1.16

Add:

• Add __add__ and __mul__ to space.Vector;
• model.MLModel.predict now returns the model error rate.

Fix:

• model.Model is now model.MLModel;
• parser.kfolds_split has been deprecated and removed;
• model.MLModel.cross_val_predict now uses sklearn.model_selection.StratifiedKFold for the generation of balanced folds;
• Fix the order of the test real labels before computing the model metrics in examples/chopin2.py.

hdlib v0.1.15

hdlib v0.1.15

Add:

• Add examples/chopin2_iris.sh as a test case for examples/chopin2.py;
• Add new unit tests to test/test.py.

Fix:

• space.Space.bulk_insert has been refactored to make use of space.Space.insert;
• parser.load_dataset now throws a ValueError in case of non-numerical datasets;
• Add missing import os in space.Model.

hdlib v0.1.14

hdlib v0.1.14

Fix:

• model.Model.fit now correctly generates both bipolar and binary level vectors;
• space.Vector.dist automatically converts the cosine similarity into a distance measure;
• model.Model.predict and model.Model.error_rate are now compatible with all the supported distance metrics (euclidean, hamming, and cosine).

hdlib v0.1.13

hdlib v0.1.13

Fix:

• Fix the retraining process in model.Model.predict.

hdlib v0.1.12

hdlib v0.1.12

Add:

• examples/chopin2.py now reports the Accuracy, F1, Precision, Recall, and the Matthews correlation coefficient for each of the folds in addition to the average of these scores as evaluation metrics of the hyperdimensional computing models;
• model.Model class functions now raise different exceptions based on multiple checks on the input parameters.

hdlib v0.1.11

hdlib v0.1.11

Fix:

• The model.Model.stepwise_regression function now report the importance corresponding to the best score;
• The model.Model._init_fit_predict function uses average="weighted" for computing a score different from the accuracy to account for label imbalance;
• examples/chopin2.py now computes different scores on the resulting predictions, prints the list of selected features based on the best score, and finally reports the confusion matrices.

hdlib v0.1.10

hdlib v0.1.10

Add:

• Add error_rate as model.Model class method for computing the error rate of a classification model.

Fix:

• The model.Model.predict function computes the error rate before retraining the classification model.

hdlib v0.1.9

hdlib v0.1.9

Fix:

• Fix the retrining process in model.Model.predict to avoid overfitting.

⚠️ Avoid using previous versions of hdlib.