=== Library ===
==== Major changes ====
* PCE: polynomial cached evaluations
* Kriging: new kernels including anisotropic ones
* Distribution: more efficient algebra, more copulas and multivariate distributions
* Bayesian modeling: improved MCMC, BayesDistribution, enhanced ConditionalDistribution, conjugate priors for Normal distribution
==== New classes ====
* AggregatedProcess, allowing to stack processes with common spatial dimension
* ProductDistribution class, dedicated to the modeling of the distribution of the product of two independent absolutely continuous random variables.
* MaximumEntropyStatisticsDistribution
* MaximumEntropyStatisticsCopula
* CovarianceHMatrix, which can be used by TemporalNormalProcess to approximate covariance matrix via an H-Matrix library.
* InverseChiSquare
* InverseGamma
* NormalGamma
* OrdinalSumCopula
* MaternModel
* ProductCovarianceModel
* BoxCoxGradientImplementation
* BoxCoxHessianImplementation
* InverseBoxCoxGradientImplementation
* InverseBoxCoxHessianImplementation
* KrigingResult
* BayesDistribution
* PythonNumericalMathGradientImplementation
* PythonNumericalMathHessianImplementation
* PythonDynamicalFunctionImplementation
==== API changes ====
* Deprecated method NumericalMathFunction|NumericalMathFunctionEvaluation::getOutputHistory|getInputHistory in favor of NumericalMathFunction::getHistoryOutput|getHistoryInput
* Removed method Graph::initializeValidLegendPositions
* Renamed the getMarginalProcess() method into getMarginal() in the Process class and all the related classes.
* Deprecated methods Graph::getBitmap|getPostscript|getVectorial|getPath|getFileName
* Deprecated methods Graph::draw(path, file, width, height, format), use draw(path+file, width, height, format) instead
* Removed deprecated methods ResourceMap::SetAsUnsignedLong|GetAsUnsignedLong in favor of ResourceMap::SetAsUnsignedInteger|GetAsUnsignedInteger
* Removed deprecated methods NumericalSample::scale|translate
* Renamed the acosh(), asinh(), atanh() and cbrt() methods of the SpecFunc class into Acosh(), Asinh(), Atanh() and Cbrt() and provided custom implementations.
* Added the rUniformTriangle() method to the DistFunc class to generate uniform random deviates in a given nD triangle.
* Extended the GaussKronrod, IntegrationAlgorithm and IntegrationAlgorithmImplementation classes to multi-valued functions.
* Extended the FFT and RandomMixture classes to 2D and 3D.
* Added the setValues() method to the Field class.
* Added Simulation::setProgressCallback|setStopCallback to set up hooks
* Added the getParameterDimension() method to the NumericalMathFunction class.
* Added new parallel implementations of the discretize() and discretizeRow() methods in the CovarianceModelImplementation class.
* Added the key "Os-RemoveFiles" to the ResourceMap class.
* Added the BesselK(), LogBesselK() and BesselKDerivative() methods to the SpecFunc class.
* Added the spatial dimension information to the CovarianceModel class.
* Added a discretize() method based on sample to the CovarianceModel class.
* Added a nugget factor to all the covariance models.
* Added an history mechanism to the MCMC class.
* Added accessors to the amplitude, scale, nugget factor, spatial correlation to the CovarianceModel class.
* Added the getLogLikelihoodFunction() method to the KrigingAlgorithm class.
* Added a link function to the ConditionalDistribution class.
* Added the getMarginal(), hasIndependentCopula(), hasEllipticalCopula(), isElliptical(), isContinuous(), isDiscrete(), isIntegral() methods to the RandomMixture class.
* Added the getSupport() and the computeProbability() methods to the Mixture class.
* Added a simplified constructor to the BayesDistribution class.
* Added the computeRange() and getMarginal() methods to the BayesDistribution class.
* Added the isIncreasing() method to the Indices class.
* Added a dedicated computeLogPDF() method to the Rice class.
* Added the LargeCaseDeltaLogBesselI10() and DeltaLogBesselI10() methods to the SpecFunc class.
* Removed the useless getPartialDiscretization() method to the CovarianceModel class.
* Removed the getConditionalCovarianceModel() in the KrigingAlgorithm class.
* Renamed the getMeshDimension() method into getSpatialDimension() in the DynamicalFunction class.
* Renamed the isNormal(), isInf() and isNaN() methods into IsNormal(), IsInf() and IsNan() in the SpecFunc class.
* Removed FittingTest::GetLastResult, FittingTest::BestModel*(sample, *) in favor of FittingTest::BestModel*(sample, *, &bestResult)
* Deprecated NumericalMathFunction(Implementation)::set{Evaluation|Gradient|Hessian}Implementation in favor of NumericalMathFunction(Implementation)::set{Evaluation|Gradient|Hessian}
* Deprecated NumericalSample::compute{Range,Median,Variance,Skewness,Kurtosis,CenteredMoment,RawMoment}PerComponent
* Deprecated ProcessSample::setField(index, field) in favor of ProcessSample::setField(field, index)
=== Python module ===
* Include sphinx documentation
* Improved collection accessors
* Allow one to overload gradient and hessian
* Improved viewer's integration with matplotlib api
* Added PythonDynamicalFunction to override DynamicalFunction
=== Miscellaneous ===
* In Graph::draw, the file extension overrides the format argument
* Improved the compactSupport() method of the UserDefined class. Now, it works with multidimensional distributions.
* Improved the computePDF() and computeCDF() methods of the UserDefined class.
* Improved the RandomMixture class to allow for constant distribution and Dirac contributors.
* Added /FORCE option to windows installer to allow out-of-python-tree install
* Added a generic implementation of the getMarginal() method to the Process class for 1D processes.
* Added a description to all the fields generated by a getRealization() method of a process.
* Changed the values of the keys ConditionalDistribution-MarginalIntegrationNodesNumber, KernelSmoothing-BinNumber, SquaredExponential-DefaultTheta, AbsoluteExponential-DefaultTheta, GeneralizedExponential-DefaultTheta in the ResourceMap class and the openturns.conf file.
* Changed the parameterization of the AbsoluteExponential, GeneralizedExponential and SquaredExponential classes.
* Changed the default parameterization of the ComposedCopula, ConditionalDistribution, AliMikhailHaqCopula, FarlieGumbelMorgensternCopula, KernelMixture, Mixture and NormalCopula classes.
* Changed the default presentation of analytical functions.
* Changed the parameters of the default distribution of the FisherSnedecor class.
* Changed the algorithm used in the FisherSnedecorFactory class. Now the estimation is based on MLE.
* Extended the Debye() method of the SpecFunc class to negative arguments.
* Extended the computeCDF(), computeDDF(), computeProbability() methods of the RandomMixture class.
* Extended the ConditionalDistribution class to accept a link function.
* Extended the build() method of the IntervalMesher class to dimension 3.
* Improved the capabilities of the KrigingAlgorithm class. Now it can use anisotropic covariance models.
* Improved the __str__() method of the CompositeDistribution class.
* Improved the numerical stability of the computeCharacteristicFunction() in the Beta class.
* Improved the distribution algebra in the DistributionImplementation class.
* Improved the getKendallTau() and computeCovariance() methods of the SklarCopula class.
* Improved the Gibbs sampler in the TemporalNormalProcess class.
* Improved the presentation of the graphs generated by the drawPDF() and drawCDF() methods of the distributions.
* Improved the messages sent by the NotYetImplementedException class.
* Improved the pretty-print of the NumericalMathFunction class.
* Improved the HistogramFactory and KernelSmoothing classes by using inter-quartiles instead of standard deviations to estimate scale parameters.
* Improved the management of small coefficients in the DualLinearCombinationEvaluationImplementation class.
* Improved the algorithms of the getRealization() and computePDF() methods of the Rice class.
* Improved the operator() method of the PiecewiseLinearEvaluationImplementation class.
=== Bug fixes ===
* 614 (FORM Method - Development of sensitivity and importance factors in the physical space)
* 673 (Perform the computeRange method of the PythonDistributionImplementation class)
* 678 (Pretty-printer for gdb)
* 688 (incorrect analytical gradient)
* 704 (Problem with Exception)
* 709 (MatrixImplementation::computeQR issues)
* 713 (Dirichlet hangs on np.nans)
* 720 (Missing LHSExperiment::getShuffle)
* 721 (Python implementation of a NumericalMathGradientImplementation)
* 731 (Problems with Rice and FisherSnedecor distributions)
* 736 (Graph : keep getBitmap, getVectorial, getPDF, getPostScript, initializeValidLegendPositions?)
* 737 (Bug in composeddistribution inverse iso-probabilistic transformation in the ellipical distribution case )
* 738 (Incorrect pickling of ComposedDistribution with ComposedCopula)
* 739 (Bug in the SpecFunc::LnBeta() method)
* 744 (Incorrect iso-probabilistic transformation for elliptical ComposedDistribution)
* 745 (DirectionalSampling: ComposedCopula bug and budget limitation ignored)
* 747 (Packaging for conda)
* 748 (Can't add sklar copula to CopulaCollection)
* 754 (Bad conversion list to python with negative integer)
* 755 (inconsistency in functions API)
* 757 (Spearman correlation in CorrelationAnalysis)
* 759 (Problem with RandomMixture::project)
* 762 (NumericalSample's export produce empty lines within the Windows environment)
* 763 (Missing description of samples with RandomVector realizations)
* 764 (RandomVector's description)
* 769 (Dirichlet behaves strangely on constant)
* 770 (Problem with FittingTest based on BIC)