v0.0.13

Improve

• The eigenvalue calculation of 3*3 matrix to be done algebraically
• Add tolerance parameter to some method of Matrix
• Add isAlternating and isSkewHermitian method in Matrix class
• Add tridiagHouseholder, tridiagLanczos, hessenberg and hessenbergArnoldi method in Matrix class
• Improve eigenValuesQR method in Matrix class
• Increase value of batch_size in DQN class
• Add detach layer of neuralnetwork
• Add A2C model

Bug fix

• isIdentity and isRegular functions of Matrix class returns incorrect result
• 0 becomes null by negative function of Matrix class
• Fix randn function of Matrix class

v0.0.12

Improve

• Add association analysis
• Add argmax, RReLU, RNN, LSTM and GRU layer of neuralnetwork
• Use momentum to learning hyperparameter of PReLU
• Not use cosh to calculate the gradient of tanh
• Remove id parameter of include layer
• Add equals method in Matrix and Tensor class
• Fill 0 in empty element of Matrix class
• Change separator of toString of Tensor class

Bug fix

• removeAt affects an array passed to the constructor of Tree class
• Failed to use some layer
• Gradient of leaky ReLU and PReLU layer is not correct
• sampleCol fails when index is true
• qr function returns incorrect result when the number of columns is 1
• eigenValuesQR failes when the matrix size is less than 2

v0.0.11

Improve

• Add breaker (means 'breakout') reinforcement environment
• Split file of neuralnetwork layers
• Add neuralnetwork layers (ELU, PReLU and Softargmax)
• Add hyperparameter of softplus layer
• Improve softmax layer
• Add functions (removeRowIf, removeColIf, every and some) in Matrix class

Breaking change

• Remove polynomial layer (which is same as power layer)

Bug fix

• Binary operations and cholesky decomposition of Matrix does not returns correct result
• LU decomposition returns only one value when the Matrix size is zero
• Return value of toArray in Matrix class contains empty value which means 0
• reduce returns incorrect result when the type of initial value is boolean
• Gradient of many layers is incorrect

v0.0.10

Improve

• Add Tukey's fences anomaly detection

Bug fix

• Incorrect percentile calculation

v0.0.9

Breaking change

• Change function name from exponentialMovingAverate to exponentialMovingAverage
• Change arguments of the fit function of Autoencoder class

Document

• Add jsdoc

v0.0.8

Improve

• Add ID3 method in decision tree and random forest
• Export reinforcement learning environment of games

Document

• Create document

v0.0.7

Improve

• Export reinforcement learning environment

v0.0.6

Improve

• Export RuLSIF, uLSIF model

Bug fix

• The random number generated from the normal distribution is incorrect.
• The results of the Cholesky decomposition are not correct.

v0.0.5

Bug fix

• There are some untracked classes

v0.0.4

Bug fix

• index.js become empty