1.5-beta.3

Additions

Package dynaml.probability

• Added method KL() to probability package object, to calculate the Kulback Leibler divergence between two continuous random variables backed by breeze distributions.

Improvements

Package dynaml.probability

• Changes to RandomVarWithDistr: made type parameter Dist covariant.
• Reform to IIDRandomVar hierarchy.

1.5-beta.2

Additions

• Added installation script install.sh
• Added ggplot2 to renjin package dependencies.

Package dynaml.probability

• Added ContMixtureRVBars a private class which can be instantiated using the apply method of ContinuousDistrMixture

/*
The following import is needed for an implicit parameter
which serves as the inner product space on the domain 
i.e. Inner products on Double
*/
import spire.implicits._

//Initialise the mixture components
val components = Seq(UnivariateGaussian(0.0, 1.0), UnivariateGaussian(-1.5, 0.2))

val gaussian_mixture = ContinuousDistrMixture[Double, Double, UnivariateGaussian](
  components, DenseVector(0.65, 0.35))

Package dynaml.models

• Added API starting points for implementations of stochastic mixture models. The key classes/traits are
  • StochasticProcessMixtureModel: The base class
  • ContinuousMixtureModel: An abstraction for mixture models building on subtypes of ContinuousProcessModel as the base processes. Offers implementation of only the predictiveDistribution() method.
  • GenContinuousMixtureModel: This builds mixture models based on stochastic processes which return predictive distributions in closed form, having mean, variance and error bars. Implements all methods except toStream(y) and getVectorSpace(num_dim) which are left to the user.

Package dynaml.models.gp & dynaml.models.stp

• Added classes GaussianProcessMixture, StudentTProcessMixture and MVTMixture representing stochastic mixtures over gaussian processes, student t processes and matrix variate t processes respectively.

Package dynaml.optimization

• Added abstract class MixtureMachine which takes as input a stochastic process model and returns a stochastic mixture model with weights computed using a grid search or coupled simulated annealing procedure.

• Added GPMixtureMachine an extension of MixtureMachine as a convenience class for creating gaussian process stochastic mixtures.

Package dynaml.kernels

• Added separable stationary kernel implementation in SeparableStationaryKernel
  as specified by Genton et. al.

1.5-beta.1

Additions

Package dynaml.algebra

• Added support for dual numbers.

  //Zero Dual
  val zero = DualNumber.zero[Double]  
  
  val dnum = DualNumber(1.5, -1.0) 
  val dnum1 = DualNumber(-1.5, 1.0) 
  
  //Algebraic operations: multiplication and addition/subtraction
  dnum1*dnum2
  dnum1 - dnum
  dnum*zero 

Package dynaml.probability

• Added support for mixture distributions and mixture random variables. MixtureRV, ContinuousDistrMixture for random variables and MixtureDistribution for constructing mixtures of breeze distributions.

Package dynaml.optimization

• Added ModelTuner[T, T1] trait as a super trait to GlobalOptimizer[T]
• GridSearch and CoupledSimulatedAnnealing now extend AbstractGridSearch and AbstractCSA respectively.
• Added ProbGPMixtureMachine: constructs a mixture model after a CSA or grid search routine by calculating the mixture probabilities of members of the final hyper-parameter ensemble.

Package dynaml.models

• Added StochasticProcessMixtureModelas top level class for stochastic mixture models.
• Added GaussianProcessMixture: implementation of gaussian process mixture models.

Fixes

Package dynaml.optimization

• Corrected energy calculation in CoupledSimulatedAnnealing; added log likelihood due to hyper-prior.

1.4.3

DynaML REPL

Additions

Module scripts

• Added gp_mcmc_santafe.sc worksheet to try new MCMC feature on GP models; applied on the Santa Fe laser data set.

General

• Updated Ammonite version to 0.9.9

Pipes API

Additions

Package dynaml.pipes

• Added ._1 and ._2 members in ParallelPipe

Core API

Additions

Package dynaml.models.neuralnets

• Added SELU activation function proposed by Hochreiter et. al

Package dynaml.models.bayes

• Added * method to CoRegGPPrior which scales it with to a ParallelPipe

Package dynaml.probability.mcmc

• Added HyperParameterMCMC for performing MCMC sampling for models extending GloballyOptimizable.

Package dynaml.utils

• Added trait HyperParameters outlining methods that must be implemented by entities having hyper-parameters

• Added MeanScaler, PCAScaler to perform mean centering and PCA on data sets. Also added to DynaMLPipe pipe library.

• Added tail recursive computation of the Chebyshev polynomials of the first and second kind in utils.chebyshev.

Improvements

Package dynaml.models.bayes

• Added trendParamsEncoder which converts the trend/mean parameters into a scala Map[String, Double] making them
  consistent with covariance parameters. Added to GaussianProcessPrior and ESGPPrior families.

1.4.3-beta.3

Additions

Package dynaml.probability.mcmc

• Added HyperParameterMCMC for performing MCMC sampling for models extending GloballyOptimizable.

Package dynaml.utils

• Added trait HyperParameters outlining methods that must be implemented by entities having hyper-parameters

• Added MeanScaler, PCAScaler to perform mean centering and PCA on data sets. Also added to DynaMLPipe pipe library.

Module scripts

• Added gp_mcmc_santafe.sc worksheet to try new MCMC feature on GP models; applied on the Santa Fe laser data set.

1.4.3-beta.2

Improvements

Package dynaml.models.bayes

• Added trendParamsEncoder which converts the trend/mean parameters into a scala Map[String, Double] making them consistent with covariance parameters. Added to GaussianProcessPrior and ESGPPrior families

1.4.3-beta.1

Additions

Package dynaml.pipes

• Added ._1 and ._2 members in ParallelPipe

Package dynaml.models.bayes

• Added * method to CoRegGPPrior which scales it with to a ParallelPipe

1.4.2

Additions

Package dynaml.models.neuralnets

• Added GenericAutoEncoder[LayerP, I], the class AutoEncoder is now deprecated
• Added GenericNeuralStack[P, I, T] as a base class for Neural Stack API
• Added LazyNeuralStack[P, I] where the layers are lazily spawned.

Package dynaml.kernels

• Added ScaledKernel[I] representing kernels of scaled Gaussian Processes.

Package dynaml.models.bayes

• Added * method to GaussianProcessPrior[I, M] which creates a scaled Gaussian Process prior using the newly minted ScaledKernel[I] class
• Added Kronecker product GP priors with the CoRegGPPrior[I, J, M] class

Package dynaml.models.stp

• Added multi-output Students' T Regression model of Conti & O' Hagan in class MVStudentsTModel

Package dynaml.probability.distributions

• Added HasErrorBars[T] generic trait representing distributions which can generate confidence intervals around their mean value.

Improvements

Package dynaml.probability

• Fixed issue with creation of MeasurableFunction instances from RandomVariable instances

Package dynaml.probability.distributions

• Changed error bar calculations and sampling of Students T distributions (vector and matrix) and Matrix Normal distribution.

Package dynaml.models.gp

• Added Kronecker structure speed up to energy (marginal likelihood) calculation of multi-output GP models

Package dynaml.kernels

• Improved implicit paramterization of Matern Covariance classes

General

• Updated breeze version to latest.
• Updated Ammonite version to latest

1.4.2-beta.4

Additions

Package dynaml.models.stp

• Added multi-output Students' T Regression model of Conti & O' Hagan in class MVStudentsTModel

Package dynaml.probability.distributions

• Added HasErrorBars[T] generic trait representing distributions which can generate confidence intervals around their mean value.

Improvements

Package dynaml.probability.distributions

• Changed error bar calculations and sampling of Students T distributions (vector and matrix) and Matrix Normal distribution.

General

• Updated breeze version to latest.

1.4.2-beta.3

Improvements

• Refined implementation of GenericNeuralStack and its hierarchy.