v0.7.1

Big fix

• Fix reshape layer #101

Full Changelog: v0.7.0...v0.7.1

v0.7.0

Improve

• Add some norm function in Matrix class #94

Bug fix

• Fix Markov switching #86
• Fix CumSum #88

Minor changes

• Add tests #91
• Fix type declaration document for evaluate methods #92

Full Changelog: v0.6.0...v0.7.0

v0.6.0

Improve

• Add S3VM #74

Bug fix

• Fix Otus's thresholding #78

Breaking changes

• Changed the value returned by the binarization method to 0 or 1. #77
• Improve matrix class #82

Minor changes

• Add test for split and merge #80

Full Changelog: v0.5.0...v0.6.0

v0.5.0

Improve

• Add Weighted k-means #66
• Add Genetic k-means #68

Breaking changes

• Remove activation layer #60

Bug fix

• Fix principal curve #70

Full Changelog: v0.4.0...v0.5.0

v0.4.0

Improve

• Add K-SVD and improve Matrix #46
• Add sparsity parameter in MOD #49
• Add DENCLUE #53

Minor changes

• Remove Tree dependencies #58

Full Changelog: v0.3.0...v0.4.0

v0.3.0

Improve

• Add extra trees #28
• Add Slice sampling #34
• Make it possible to choose the method of AR model #36
• Add Monothetic Clustering #38
• Add HLLE #40

Breaking changes

• Remove init_args from EnsembleBinaryModel #43

Minor changes

• Remove dependency of Tree class in IsolationForest #30
• Use DFT in lowpass filter #32
• Use substring instead of substr #42

Full Changelog: v0.2.0...v0.3.0

v0.2.0

Improve

• Change random layer to be able to get sizes from other layer #4
• Change full, reshape and variable layer to accept layer names #6
• Add Ladder network #8
• Add Theil-Sen regression #20
• Add HDBSCAN #23

Bug fix

• Test the layer directly and fix some returns of toObject #18
• Fix exports path #25

Minor changes

• Generalize the genetic algorithm #12

Others

• Improve github actions #2
• Add issues and PR templates #14
• Fix issue template #16

Full Changelog: v0.1.0...v0.2.0

v0.1.0

Improve

• Add iteration, isNilpotent, exp and log function in Matrix class
• Generate Typescript type declaration files
• Add MOD dimensionality reduction

Breaking change

• Do not run tridiag or hessenberg in eigenValuesQR of Matrix class
• Split Tensor class from math.js and rename math.js to matrix.js
• Simplify k-means class structure

Bug fix

• LARS in Lasso throws error
• Some function of RLIntRange is incorrect
• reset function of EmptyRLEnvironment returns function
• The calculation of the normalization term of MultinomialLogisticRegression is incorrect
• states of BreakerRLEnvironment returns incorrect values

Minor change

• Improve comment in index.js
• Change some error classes to inherit from Error
• Change the file structure of the neuralnetwork

v0.0.26

Improve

• Add Categorical naive bayes
• Add Bi-section to calculate eivenvalues in Matrix class
• Improve error handring of some function of Matrix class

Breaking changes

• Change I/F of edge detection models
• Only accept 2D kernel to convolute function of Matrix class
• Change LDA to a class

Bug fix

• LVQ 3 failed in sometime
• SVM calculation is not correct
• fit function of k-prototypes returns undefined
• test function of Maze environment returns uncorrect values
• ICA calculation is not correct

Minor changes

• Remove many unused variables
• Improve simple layer calculation function
• Improve JSDoc

v0.0.25

Improve

• Add predict function in Histogram and AverageShiftedHistogram
• Add n-cubic interpolation
• Add Co-training algorithm
• Add Statistical Region Merging
• Add PLSA
• Additive coupling layer now can run toObject
• Add predict function to NeuralNetwork class
• Add RANSAC

Breaking changes

• Make Histogram and AverageShiftedHistogram to default export and change AverageShiftedHistogram to class object
• Rename HermitInterpolation to CubicHermiteSpline
• Change interface of Self-training
• Adaptive, Bernsen, Niblack, Phansalkar and Sauvola Thresholding accept only 2D (only have single channel) data
• predict function of SplitAndMerge returns 2D data
• Split Complex and Tree class to other file
• Change some layers to be generated by a common process

Bug fix

• Draughts board can set piece anywhere
• predict function of PCR returns invalid values

Minor changes

• Improve process of n-linear interpolation