1.4-beta.26

Improvements

1. Faster computation of kernel matrices and marginal likelihood computation for GP models

1.4-beta.25

Bugfix

1. Fix to CompositeCovariance

Release 1.4-beta.24

Bugfixes

MOGPRegressionModel[I]: mistake in dimensions of kernel matrix fixed.

Improvements

Added simple rejection sampling for posterior in ProbabilityModel

Release 1.4-beta.23

Bugfix

1. Ensured symmetric posterior covariance matrix in GP implementation.

Release 1.4-beta.22

Bug fixes

1. Introduced blocking of hyper parameters for CompositeCovariance[Index]
2. Improvements to dataAsSeq method of MOGPRegressionModel

Release 1.4-beta.21

BugFix

1. Corrected expression for evaluation of RationalQuadraticKernel.

Release 1.4-beta.20

Additions

1. Added RandomVariable and ProbabilityModel APIs to facilitate generative models based on conditional probabilities.
2. Added Wavelet class
3. WIP: Multi output GP models

Release 1.4-beta.19

Improvements

1. Added Kullback Leibler divergence to BackPropagation implementation
2. Optimized loops in solvers using cfor macro in the spire library
3. Added unit tests for DataPipe

Release 1.4-beta.18

This release contains some important additions/improvements.

Improvements/Debugging

Improvement of back-propagation implementation with respect to convergence as well as faster execution.

Enhancements

First unit tests written for the following components

Optimization solvers:

1. Gradient Descent
2. Quasi-Newton BGFS
3. Back-propagation with momentum and regularization
4. Conjugate Gradient

Models

1. Generalized Linear Models
2. Feed forward neural networks
3. Autoencoders (WIP)

Release 1.4-beta.17

Fix

1. To AutoEncoder class implementation of encoding and decoding