v0.23.0

Improve

• Accept object for add method of ComputationalGraph class #902
• Add some singular / eigen value calculate method #904
• Not print undefined of toDot function #906
• Make the I/F of the Matrix and Tensor classes the same #913
• Add a size argument to the input layer #916
• Export size via toObject method #918
• Skip adding parents if not needed #922
• Add where node importer #927
• Add ONNX export #928
• Sort and calculate according to dependencies #933
• Add as constant layer if input is initializer #934
• Add LRN layer support for ONNX export #936

Bug fix

• Fix constructor of APL layer #920
• Fix comparison layers and update related tests #938

Minor changes

• Use object literals instead of switch statements in some constructors #908
• Create document using typedoc #910
• Move type-only imports to comments #924
• Change the convergence check of eigenJacobi #930
• Improve tests #941
• Update version to 0.23.0 #942

Full Changelog: v0.22.0...v0.23.0

v0.22.0

Improve

• Add duplicate argument in sample method #893
• Add MARS #897

Minor changes

• Use docker compose command #895
• Improve tests #899
• Update version to 0.22.0 #900

Full Changelog: v0.21.0...v0.22.0

v0.21.0

Improve

• Add some neuralnetwork optimizers #863
• Add inverse function for transform and scaling models #865
• Add some neuralnetwork optimizer #871
• Add ELM #875
• Add DTSCAN #877
• Add DOC #879
• Add some iterative method of solve in Matrix class #881
• Add DiSH #885

Small changes

• Changed to use float16 instead of int16 #873
• Increase retry time of some tests #890
• Update version to 0.21.0 #891

Full Changelog: v0.20.1...v0.21.0

v0.20.1

Bug fix

• Fix index out of bounds of ORCLUS #830

Minor changes

• Remove unreachable code and unused variables #823
• Improve JSDoc #825
• Use pyenv to create onnx files #829
• Remove dependency with LowpassFilter #832
• Delete unnecessary empty line from JSDoc #842
• Update libraries #846
• Change dynamic import in library section to static import #850
• Increase retry time of some tests #854

Full Changelog: v0.20.0...v0.20.1

v0.20.0

Improve

• Add RankNet #769
• Add string expression function layer #779
• Add k-harmonic means #799
• Add some variation of k-means algorithm #801
• Accept metric as function #816

Breaking changes

• Simplify MLP implementation #775
• Accept object for kernel argument #815
• Move arguments related to learning from the predict function #818

Bug fix

• Fix ELU gradient #774
• Fix acosh gradient of function layer #803

Minor changes

• Update version of node in actions #771
• Fix JSDoc of MLP #777
• Move Dockerfile to test directory #781
• Improve tests #813
• Remove KMeans model dependency from some models #820
• Update version to 0.20.0 #821

Full Changelog: v0.19.0...v0.20.0

v0.19.0

Improve

• Add Naive bayes regression #711
• Accept activation layer as object #713
• Accept string subscript to handle optional output of onnx #719
• Improve RNN layers #722
• Add balancing method in matrix class #724
• Add maxIteration argument for some eigenValues calculation methods #726
• Add PRank #734
• Add OAP-BPM #736
• Add GNN layers #739
• Add estimate method argument for ZIP #745
• Add BRIDGE clustering #747
• Add ZINB #757
• Add NOF #764

Breaking changes

• Add ordered logistic regression and rename ordered probit model #732

Bug fix

• Fix threshold initialization in PRank and OAPBPM #741
• Fix RBP #751
• Set the default predicted value at the beginning of the fit in CLUES #755

Minor changes

• Use esbuild instead of terser for minification #720
• Estimate bias parameter in least square model #730
• Remove unused function in OAPBPM and small changes #738
• Simplify the gradient calculation for ordered logistic regression #749
• Changed to a different method of generating arrays using binary numbers #753
• Add JSDoc of MulticlassBSGD #759
• Improve tests #761
• Estimate bias parameter in LTS #766
• Update version to 0.19.0 #767

Full Changelog: v0.18.0...v0.19.0

v0.18.0

Improve

• Add some layers #677
• Add LOESS #681
• Add Multiclass Ridge classifier #697
• Add ordinal regression #705

Breaking changes

• Improve graph #673
• Move h parameter of KDE from fit method to constructor #692
• Make index method in tensor class #706

Minor changes

• Ignore coverage of onnx_pb.js #679
• Remove JSDoc of lastPath #683
• Use common layer builder for identity-like layer #686
• Improve test #689
• Skip outside calculation in maxpool layer #691
• Update github actions #701
• Update devDependencies #708
• Update version to 0.18.0 #709

Full Changelog: v0.17.0...v0.18.0

v0.17.0

Improve

• Improve NNs layers #652
• Add ENaN #656
• Improve NNs layers #659
• Implement Draughts environment #662
• Improve matrix class #664

Breaking changes

• Change GERU layer name to GELU #660
• Use ONNX proto instance #666
• Fix and improve RL environments, and add tests #668

Minor changes

• Improve tests #650
• Fix onnx_importer #654
• Fix toString of draughts board #670
• Update version to 0.17.0 #671

Full Changelog: v0.16.1...v0.17.0

v0.16.1

Improve

• Improve some functions in util #647

Minor improve

• Use pytorch for CPU #645

Full Changelog: v0.16.0...v0.16.1

v0.16.0

Improve

• Convert data types in layers #634
• Add clone method in RL environment and fix A2C model #636

Breaking changes

• Remove log space range #630

Small changes

• Use playwright and get coverage of GUI test #628

Full Changelog: v0.15.0...v0.16.0