Skip to content

Latest commit

 

History

History
259 lines (154 loc) · 10.5 KB

CHANGELOG.md

File metadata and controls

259 lines (154 loc) · 10.5 KB

Changelog

Version 2.1.0

@cumbof/hdlib@2.1.0

New features

  • Space and Vector objects are not constrained on a 10,000 dimensionality anymore;
  • The predict function in hdlib.model.classification.ClassificationModel now returns the actual distances between the test samples and the class prototypes too;
  • The classical arithmetic has been moved to hdlib.arithmetic;
  • hdlib.arithmetic.quantum provides the quantum arithmetic at the base of Quantum Hyperdimensional Computing;
  • The hdlib.model.classification module now contains a new QuantumClassificationModel class;
  • New reasoning and classification examples about the Quantum Hyperdimensional Computing are available under the examples folder.

Version 2.0.0

@cumbof/hdlib@2.0.0

New features

  • The Vector class is now in hdlib.vector;
  • hdlib.model.MLModel is now hdlib.model.classification.ClassificationModel;
  • hdlib.graph.Graph is now hdlib.model.graph.GraphModel;
  • The hdlib.parser module has been suppressed and its functions have been moved to examples/chopin2/chopin2.py;
  • New hdlib.model.graph.GraphModel._error_rate and hdlib.model.graph.GraphModel._predict static methods for the estimation of the model error rate and the evaluation of a graph model in multiprocessing;
  • New hdlib.model.regression.RegressionModel and hdlib.model.clustering.ClusteringModel classes for building regression and clustering models respectively;
  • New examples about the classification of chemical structures and the classification of viral species via graph encoding.

Fixes

  • hdlib.model.graph.GraphModel.error_mitigation does not take into account for false positives;
  • hdlib.model.graph.GraphModel.error_rate and hdlib.model.graph.GraphModel.predict now check whether an edge exist using weight-specific thresholds.

Version 0.1.20

@cumbof/hdlib@0.1.20

Fixes

  • Replace numpy.NINF with -numpy.inf as a change introduced in Numpy 2.0.

Version 0.1.19

@cumbof/hdlib@0.1.19

Fixes

  • Replace numpy.Inf with numpy.inf as a change introduced in Numpy 2.0;
  • hdlib.graph and the corresponding unittest are now working properly.

Version 0.1.18

@cumbof/hdlib@0.1.18

New features

  • 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.

Fixes

  • 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.

Version 0.1.17

@cumbof/hdlib@0.1.17

New features

  • 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.

Version 0.1.16

@cumbof/hdlib@0.1.16

New features

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

Fixes

  • 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/chopin2.py.

Version 0.1.15

@cumbof/hdlib@0.1.15

New features

Fixes

  • 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.

Version 0.1.14

@cumbof/hdlib@0.1.14

Fixes

  • 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).

Version 0.1.13

@cumbof/hdlib@0.1.13

Fixes

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

Version 0.1.12

@cumbof/hdlib@0.1.12

New features

  • examples/chopin2/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.

Version 0.1.11

@cumbof/hdlib@0.1.11

Fixes

  • 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/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.

Version 0.1.10

@cumbof/hdlib@0.1.10

New features

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

Fixes

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

Version 0.1.9

@cumbof/hdlib@0.1.9

Fixes

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

⚠️ Avoid using previous versions of hdlib.

Version 0.1.8

@cumbof/hdlib@0.1.8

Fixes

  • Fix the initialization of Vector objects with a specific seed;
  • model.Model._init_fit_predict and model.Model._stepwise_regression_iter are now private;
  • Improving docstring adopting the numpydoc documentation format.

Version 0.1.7

@cumbof/hdlib@0.1.7

Fixes

  • Fix the break condition in model.Model.stepwise_regression for both the backward and forward methods.

Version 0.1.6

@cumbof/hdlib@0.1.6

New features

  • Add stepwise_regression as model.Model class method for performing the feature selection as backward variable elimination or forward variable selection.

Version 0.1.5

@cumbof/hdlib@0.1.5

New features

  • Add bind, bundle, and permute as Vector class methods for applying arithmetic operations inplace;
  • Rename split_dataset in the parser module into kfolds_split;
  • Add percentage_split function to the parser module;
  • Integrate chopin2 ML model into the model module withe the Model class.

Version 0.1.4

@cumbof/hdlib@0.1.4

New features

  • Add parser module with utility functions for dealing with input datasets;
  • Check if the input pickle file exists before initializing Vector and Space objects with from_file;
  • Report Vector and Spaceobjects information when calling print;
  • Add examples/chopin2/chopin2.py: reimplementation of the chopin2 ML model with hdlib.

Fixes

  • Tags are maintained as sets when applying arithmetic operators.

Version 0.1.3

@cumbof/hdlib@0.1.3

New features

  • Check for version compatibility when loading a pickle file;
  • Add normalize() function to the Vector class;
  • Link vectors in space with parents and children sets as class attributes;
  • Add link() and set_root() functions to the Space class.

Version 0.1.2

@cumbof/hdlib@0.1.2

New features

  • New distance metrics: cosine, hamming, and euclidean;
  • Tags are inherited after applying the arithmetic operators;
  • Unit tests for Vector and Space classes and for bundle, bind, and permute arithmetic operators;
  • What is the Dollar of Mexico? as a unit test.

Version 0.1.1

@cumbof/hdlib@0.1.1

New features

  • Vectors can be tagged;
  • Tags can be used to retrieve groups of Vectors in the Space with the get method of the Space class;
  • The performances of the remove and findAll methods of the Space class have been improved.

Version 0.1.0

@cumbof/hdlib@0.1.0

First public release of hdlib.

Hyperdimensional Computing Library for building Vector Symbolic Architectures in Python 3