Accord.NET Framework 3.8.0 release notes
-----------------------------------------
22.10.2017.

Version updates and fixes:
  - GH-82: Add support for weighted PCA;
  - GH-127: Fast KMeans (Request);
  - GH-145: MovingNormalStatistics;
  - GH-157: Issue in Survival analysis using VB.NET;
  - GH-184: Add an Example for Graylevel coocurrences;
  - GH-211: Any samples on how to use Boosted Decision Trees;
  - GH-257: DFT functions in AForge.Math.FourierTransform and Accord.Math.Transforms;
  - GH-262: C45Learning Discrete vs Real;
  - GH-374: Dictionary of video capabilities doesn't take into account the video framerate;
  - GH-376: Add an Example for VideoCaptureDevice Class;
  - GH-377: Add an Example for LevenbergMarquardt Class;
  - GH-415: Add an Example for AdaBoost(TModel);
  - GH-421: Add an Example for KalmanFilter2D Class;
  - GH-422: Add an Example for DecisionStump.Learn Method;
  - GH-424: Add an Example for AdaBoost(TModel) Class;
  - GH-430: Add an Example for GeneralConfusionMatrix Constructor (Int32, Int32[], Int32[]);
  - GH-440: BagOfWords for audio signals;
  - GH-441: Add Mel Frequency Cepstral Coefficients (MFCC);
  - GH-466: Add an Example for Distance.Mahalanobis Method (Double[], Double[]);
  - GH-467: Add an Example for Spline Class;
  - GH-473: Add an Example for IParallel.ParallelOptions Property;
  - GH-476: Add an Example for TwoWayAnova Constructor (Double[][][], TwoWayAnovaModel);
  - GH-483: Add an Example for Haralick Class;
  - GH-486: Add an Example for LibSvmModel.Save Method (String);
  - GH-498: Add an Example for QLearning Class;
  - GH-500: Add an Example for FourierTransform2.FFT Method (Complex[], FourierTransform.Direction);
  - GH-505: Add an Example for ConfusionMatrix Constructor (Int32[], Int32[], Int32, Int32);
  - GH-517: Add an Example for Sarsa Class;
  - GH-519: Add an Example for NelderMead Class;
  - GH-530: Add an Example for Matrix.Inverse Method (Double[][]);
  - GH-532: Add an Example for GaussNewton Class;
  - GH-547: Add an Example for HaralickDescriptor Class;
  - GH-554: Add an Example for BinaryTree(TNode) Class;
  - GH-557: Add an Example for Matrix.Sort(TKey, TValue) Method (TKey[], TValue[,], IComparer(TKey));
  - GH-560: Add an Example for FourierTransform2.FFT Method (Double[], Double[], FourierTransform.Direction);
  - GH-566: Add an Example for Distance.GetDistance(T) Method;
  - GH-569: Add an Example for Distance.Euclidean Method (Double, Double, Double, Double);
  - GH-575: Add an Example for LuDecomposition Class;
  - GH-576: RandomForestLearning: Examples can't run with SampleRatio not equal 1.0;
  - GH-582: Add an Example for Matrix.Multiply Method (Double[], Double[,]);
  - GH-610: Add an Example for UnivariateContinuousDistribution.Fit Method (Double[]);
  - GH-616: Add an Example for LevenbergMarquardt Class;
  - GH-618: Add an Example for Apriori Class;
  - GH-629: Add an Example for AdaBoost(TModel) Class;
  - GH-636: Add an Example for Measures.Correlation Method (Double[][]);
  - GH-640: Add an Example for GaussianKernel Class;
  - GH-642: Add an Example for Matrix.PseudoInverse Method (Decimal[,]);
  - GH-653: Add an Example for HistogramsOfOrientedGradients Class;
  - GH-656: Add an Example for MatReader.Read Method (String);
  - GH-660: Add an Example for LogLikelihoodLoss Class;
  - GH-665: Add an Example for FourierTransform.FFT Method;
  - GH-687: Add an Example for ShapiroWilkTest Class;
  - GH-695: Add an Example for TFIDF.Transform Method (String[][]);
  - GH-703: Add an Example for Imputation Class;
  - GH-717: Possible issue with DynamicTimeWarp kernel class;
  - GH-718: Add an Example for Cosine.Distance Method;
  - GH-723: Procrustes analysis is giving weird/wrong results;
  - GH-727: Add an Example for IRadialBasisKernel Interface;
  - GH-730: Binary-Split with normalized FREAK;
  - GH-739: Add an Example for MultipleLinearRegression.CoefficientOfDetermination Method;
  - GH-756: Add an Example for ProportionalHazardsAnalysis.LogLikelihood Property;
  - GH-764: Add an Example for AndersonDarlingTest Class;
  - GH-769: Issue using visual bag of words with large images;
  - GH-783: Add an Example for LocalBinaryPattern Class;
  - GH-785: Add an Example for Tools.RandomGroups Method (Int32, Double);
  - GH-787: Add an Example for HiddenMarkovModel(TDistribution, TObservation).Predict Method (TObservation[]);
  - GH-789: Add support for OS X;
  - GH-792: Add an Example for FisherExactTest Class;
  - GH-793: Add an Example for HoughLineTransformation Class;
  - GH-798: System.AccessViolationException in FastBoxBlur;
  - GH-800: Missing dependency for Accord.Neuro in NuGet;
  - GH-802: Index outside of the bounds of the array in Naive Bayes;
  - GH-803: NaN probabilities from large features with MultinomialLogisticRegression;
  - GH-805: Unsafe keyword being exposed in the public API;
  - GH-807: Add an Example for CrossValidating NaiveBayes;
  - GH-809: The Codification filter should honor the value of DefaultMissingValueReplacement unless overriden;
  - GH-811: Naive Bayes should provide better argument checking for negative symbols;
  - GH-812: ZhangSuenSkeletonization filter not exist to use;
  - GH-814: Add an Example for MulticlassSupportVectorMachine(TKernel, TInput) Class;
  - GH-818: Add an Example for LinearConstraint Class;
  - GH-819: Quadratic Objective Function to support basic vector operations;
  - GH-820: Augmented Lagrangian to support linear constraints;
  - GH-824: Improve number of class inference in ZeroOneLoss;
  - GH-825: Replace multi-dimentional with jagged arrays in IntegralImage.cs;
  - GH-828: Accord.Neuro under .Net Standard 2.0;
  - GH-830: Read PGM image pending;
  - GH-831: Index outside of the bounds of the array in Naive Bayes;
  - GH-843: Where is Accord.NET AdaBoost Decide method;
  - GH-845: Add an Example for Decision Structure;
  - GH-848: Wilcoxon Signed Rank Test for PAIRED samples: TwoSampleWilcoxonSignedRankTest;
  - GH-849: TwoSampleWilcoxonSignedRankTest crashing when sample vectors are exactly the same values;
  - GH-852: Add an Example for DecisionSet Class;
  - GH-853: Access to last Hessian in BoundedBroydenFletcherGoldfarbShanno;
  - GH-856: Add an overload to IsSymmetric that accepts a tolerance;
  - GH-857: Mann-Whitney-U Test producing strange results;
  - GH-862: Accord.Math -> Vector -> T[] Sample<T>(T[] values, int size) incorrect;
  - GH-865: Measures.Quartiles: value for Q1 (lower quartile) wrong in QuantileMethod.R;
  - GH-873: Add an Example for DecisionRule Class;
  - GH-876: Allow the maximum frame rate possible in DirectShow for VideoCaptureDevice;
  - GH-877: Add an Example for HaarCascadeWriter Class;
  - GH-878: Accord.Math.Transforms.FourierTransform2.DFT2 Is bugged;
  - GH-882: Adding lazy evaluation to matrix decompositions;
  - GH-885: Add an Example for Signal.GetEnergy Method;
  - GH-890: Add an Example for MultinomialLogisticRegressionAnalysis Class;
  - GH-897: Wrong status text in ImageView from DebugVisualizer;
  - GH-898: The Range method is producing some unexpected results;
  - GH-899: Add an Example for IntegralImage2 Class;
  - GH-900: Add an Example for ExhaustiveTemplateMatching Class;
  - GH-901: Add an Example for HSL Class;
  - GH-911: Character case of folder name;
  - GH-913: KNearestNeighbors can not be serialized;
  - GH-917: Two C++ projects require "Platform toolset v141" which is only available on VS2017;
  - GH-919: Build failed for Samples.sln on VS 2015;
  - GH-921: Fix for the normal random number generator when a seed is specified;
  - GH-924, GH-925: Fixing the seeded exponential generator;
  - GH-927: Broadcasting dimension seems counter-intuitive;
  - GH-929: Add an Example for SpeededUpRobustFeaturesDescriptor Class;
  - GH-930: Add an Example for FastCornersDetector Class;
  - GH-931: Add an Example for MatchingTracker Class;
  - GH-937: Add an Example for LogisticRegression Class;
  - GH-948: Accord.Video.FFMPEG.VideoFileReader should provide frame-based random access;
  - GH-949: Add the Free Spoken Digits Dataset to Accord.DataSets;
  - GH-950: Add a dataset for example test videos;
  - GH-955: KalmanFilter2D throws System.NullReferenceException;
  - GH-956: Integrate AForge.NET fixes (up to September 6);

* General
  - The libsonly script is now in RAR4 format instead of RAR5 so they 
    will not be listed as corrupted files by Linux/MacOSX decompressors;

* Core
  - Splitting ITransform into ITransform and ICovariantTransform (to support generic covariance);

* Video.FFMPEG
  - Standardizing the C++ projects to depend on VS2015 runtime instead of VS2017 to keep compatibility with VS2015;
  - Adding a static constructor in the FFMPEG project to check whether the system has those dependencies installed;

* Audio
  - Adding the initial version for a MelFrequencyCepstrumCoefficients audio feature extractor;
  - Adding a IAudioFeatureExtractor interface (akin to the IImageFeatureExtractor for Accord.Imaging);
  - Adding a Mono filter to convert multi-channel audio signals into single channel (mono) signals;
  - Adding a Signal.FromFile() method to load audio signals from file similarly to Bitmap.FromFile();
  - Adding an AudioDecoder class akin to ImageDecoder to find audio format decoders based on file extension;

* Audition
  - Adding BagOfAudioWords class to compute bag-of-word representations from audio signals;

* Imaging
  - Adding support for decoding PNM files in format P2 and P3 (besides the already supported P5 and P6);
  - Updating Haralick's to use the same normalization method as HOG and LBP;
  - Updating the color classes (RGB, HSL, YCbCr) to be structs;
  - Adding conversion operators between different color classes;
  - Updating ImageDecoder to find decoders using reflection instead of manual registration;
  - Updating the feature extraction framelet to implement the ITransform interfaces and deprecating IFeatureDetector;
  - Updating all feature descriptors to be classes rather than structs so the generics covariance can work;
  - Integrating AForge.NET's texture generation classes: Adding a base class for texture
    generation methods, updating them to use the framework-wide random number generator,
	and deprecating their Reset method;

* DataSets:
  - Adding the Yin-Yang dataset as an example of a non-linear 2D binary classification problem;
  - Adding the Servo dataset as an example of a mixed discrete/continuous dataset for regression;

* Math
  - Adding a methods for the numerical calculation of the Hessian in the FiniteDifferences class;
  - Updating Vector.Interval and Vector.Range to behave similar to NumPy's linspance and arange functions;
  - Adding new overloads in element-wise operations that accept a VectorType enumeration instead 
    of a integer for specifying to which dimension the element-wise operation should be performed;
  - Updating the Digamma and Trigamma functions to handle negative values;

* MachineLearning
  - Updating the AdaBoost classes to implement the more recent classification framelet;
  - Adding a Error property in ConfusionMatrix and GeneralConfusionMatrix (1.0 - Accuracy);
  - Adding named constructors to ConfusionMatrix to create matrices 
    directly from classifiers, their inputs and expected outputs;
  - Adding a new IsColor8bpp extension method to detect whether an 8-bpp image is a color image (non-grayscale);
  - Adding a new ConvertColor8bppToGrayscale8bpp extension methods to convert these into grayscale 8-bpp images;
  - Fixing Codification filter transformation for DataTables when only some columns should be converted;
  - NumberOfOutputs and NumberOfSymbols should have different implementations depending on the variable type;
  - Enforcing alphabetical/default sorted order for symbols in Codification filter (this is a breaking change);
  - Codification filter should now transform columns in the same order as specified by the user;

* Statistics
  - Adding exponentially weighted moving average (ewma) methods in Statistics.Measures partial classes;

* Sample applications
  - Adding a new sample application demonstrating how to use the framework in Unity 3D.




Accord.NET Framework 3.7.0 release notes
-----------------------------------------
20.08.2017.

Version updates and fixes:
  - GH-53: K-Medoids algorithm;
  - GH-335: Nelder-mead solver not converged;
  - GH-444: Reenable F# Unit Tests;
  - GH-587: UnmanagedImage does not supported in ExtractBiggestBlob filter;
  - GH-594: FFMPEG net35 not working;
  - GH-621: How to calculate the cache size in respect of the available RAM;
  - GH-662: 64-bit FFMPEG binaries not in output after installing with NuGet;
  - GH-669: Confusion Matrix;
  - GH-673: Stream closes after serialization with GZip compression;
  - GH-676: DoubleArrayChromosome CreateNew ignores Balancer properties;
  - GH-684: BalancedKMeans gets stuck;
  - GH-688: Cobyla constraint definitions only work with constant values;
  - GH-690: Add an Example for Cross-Validation with DecisionTrees;
  - GH-692: Add an Example for StochasticGradientDescent Class;
  - GH-693: One-class SVM decision rule;
  - GH-694: Add support for Weighted Least Squares;
  - GH-696: IndexOutOfRangeException exception in Matrix.First method;
  - GH-697: Add an Example for HiddenMarkovModel(TDistribution, TObservation);
  - GH-699: MJPEGStream throws NotImplementedException in .NET Core 2.0;
  - GH-700: MJPEGStream throws InvalidOperationException in .NET Core 2.0;
  - GH-706: DecisionTree.ToCode() returns code that does not compile;
  - GH-707: DecisionTree.ToCode() returns code that compiles;
  - GH-711: Nonlinear Regression in VB.NET;
  - GH-712: Update MJPEGStream.cs ;
  - GH-715: GeneralizedParetoDistribution shape param;
  - GH-717: Possible issue with DynamicTimeWarping kernel class;
  - GH-723: Procrustes analysis is giving weird/wrong results;
  - GH-729: Error in ExhaustiveBlockMatching;
  - GH-731: Dilatation;
  - GH-736: Measures.Quartiles() for double Vectors of size 2 is wrong;
  - GH-737: Add examples for C45Learning Class with missing data and thresholds;
  - GH-745: Cannot change degree of a default Polynomial kernel;
  - GH-746: Add an Example for CrossValidation Class;
  - GH-747: How to understand the Probabilities;
  - GH-749: 64 bit release for .NET 4.0;
  - GH-752: Speed up matrix-vector operations;
  - GH-758: NullReferenceException on NaiveBayes Learn;
  - GH-765: NaiveBayes 'System.IndexOutOfRangeException' occurred in Accord.Statistics.dll when calling from sample application;
  - GH-767: DebugVisualizers;
  - GH-777: Bug in LinearConstraintCollection documentation;
  - GH-778: Setter for bounds in BoundedBroydenFletcherGoldfarbShanno.
 
* General
  - Adding support for targetting NET Standard 1.4;
  - Adding Newtonsoft.Json (Json.NET) in externals.

* DataSets
  - Adding Wisconsin's Breast Cancer (original, prognostic and diagnostic) datasets;
  - Adding Oxford's Parkinsons dataset;
  - Updating download links for the RCV1v2 dataset.
 
* Imaging
  - Fixing multiple typos regarding the spelling of "Dilation" (this is a breaking change).
  
* IO
 - Adding a ReadLine method to CsvReader to read individual lines from the CSV file.

* MachineLearning
  - Adding K-Medoids (PAM) and Voronoi Iteration clustering algorithms;
  - Fixing epsilon in Sequential Minimal Optimization for Regression;
  - Adding a MiniBatch static class that can be used to create mini-batch definitions from training data;
  - Update LevenbergMarquardtLearning.cs to allow for different activation functions;
  - Update BackPropagationLearning.cs to allow for different activation functions;
  - Adding support for missing values in C4.5;
  - Updating GeneralConfusionMatrix to represent columns as ground-truth instead of predictions;
  - Improving memory usage for Second Order (LibSVM) Sequential Minimal Optimization;
  - Adding more overloads to the method that helps determine how many lines can 
    be included in the SVM kernel cache given a total amount of memory;
  - Fixing ToMulticlass() methods included in multi-label and binary classifiers;
  - Fixing the Probabilities and LogLikelihoods methods for multi-label and multi-class SVMs;
  - Adding an option for OneVsOneLearning/Multiclass SVMs stop at the first exception 
    found during training instead of waiting until the all machines have been trained;
  - Adding Precision, Recall, RowErrors and ColumnErrors to GeneralConfusionMatrix.
  
* Math
  - Adding support for .Learn() methods in NonlinearLeastSquares;
  - Updates GH-762: DotAndDot performance for small problem sizes;
  - Removing the deprecated extension methods for Accord.Math.Matrix.Multiply
    (such that calls should now be redirected to Elementwise.Multiply);
  - Fix BinarySearch so that it works with decreasing functions;
  - Search interval in BinarySearch was meant to be [a;b) (i.e. with inclusive a and exclusive b);
  - Fixing the behavior of Matrix.Get() method when negative indices are passed;
  - Fixing Matrix's ToTable method to use the most high level type possible when creating columns.

* Statistics
  - Adding multiple methods for computing quartiles/quantiles;
  - Adding a more advanced version of the discretization filter;
  - Adding an example for fraud detection using HMMs with MaximumLikelihoodLearning class.
  
* Neuro
  - Adding support for networks with multiple activation functions in Levenberg-Marquardt.




Accord.NET Framework 3.6.0 release notes
-----------------------------------------
09.07.2017.

Version updates and fixes:
  - GH-168: Text naive bayes classification gives wrong results;
  - GH-207: ResizeBilinear filter can support more pixel formats;
  - GH-259: K-means clustering exception;
  - GH-318: Adding support for .NET Standard 2.0;
  - GH-389: Wilcoxon Signed Rank Test / Mann-Whitney-Wilcoxon Test - differences to R;
  - GH-407: Upgrade to NUnit 3;
  - GH-470: Multiclass SVM and DTW System.AggregateException;
  - GH-499: Add an example for K-means with mixed categorical and continuous data;
  - GH-540: Add an example for BaumWelchLearning(TDistribution, TObservation, TOptions) Class;
  - GH-549: Multithreaded BagOfVisualWords: "memory corrupt" problem;
  - GH-571: Accord.Controls is referencing wrong version of ZedGraph on NuGet;
  - GH-602: Robust multivariate regression causes IndexOutOfRangeException;
  - GH-605: Renaming AudioCodec.M4A to AudioCodec.MP4ALS;
  - GH-604: ColorSlider component resource path's fix;
  - GH-606: LowerBoundNewtonRaphson: how to check if converged;
  - GH-607: from perezale/aleperez-development;
  - GH-611: Framerate issues when transcoding video;
  - GH-614: Index out of bounds error in SingularValueDecomposition.Solve;
  - GH-619: AugmentedLagrangian hangs on linear problem;
  - GH-621: How to calculate the cache size in respect of the available ram;
  - GH-624: Add an Example for Chart Class;
  - GH-627: Exposing more distribution fields as public properties;
  - GH-628: Inject other Random.Generators (like Mersenne Twister);
  - GH-630: GeneralizedBetaDistribution.ProbabilityDensityFunction does not have an area of 1;
  - GH-631: Adding test case for  GoldfarbIdnaniStatus is Success even when Solution violates constraints;
  - GH-632: ExponentialDistribution.DistributionFunction return values outside [0,1];
  - GH-647: Adding weighted versions of the Euclidean and Square Euclidean distances;
  - GH-649: Add an example for NonNegativeLeastSquares Class;
  - GH-654: Adding the Distance Transform and Watershed algorithms;
  - GH-663: Adding examples for the CsvWriter Class.

* General
  - Adding support for targetting .NET 4.6.2 and NET Standard 2.0;
  - Improving documentation and expanding number of examples.

* Core
  - Moving IParallel and ISupportsCancellation interfaces to Accord.Core.
  
* IO
  - Adding a parser for the UNIPEN file format used by the pendigits dataset;
  - Whitespace as a candidate delimiter in CSV parser;
  - Adding more overloads to SparseReader's Read method.

* DataSets
  - Renaming the previous Iris dataset to SparseIris since it was a LibSVM dataset;
  - Adding the original Iris, Wine, Pendigits, Chunking and Test Images datasets.

* MachineLearning
  - Updating Learn() methods now throw exceptions when weights are 
    passed to learning algorithms which does not yet support them;
  - Updating CrossValidation, Bootstrap, Split Set Validation and 
    Grid-Search to use the new Learn() API;
  - Adding support for creating decision trees using collection 
    initializers for the attribute/decision variables;
  - Mitigating the impact of a numerical precision issue when normalizing 
    distances to probabilities in the K-Means++ initialization;
  - Fixing issue with K-Means in which the input observations would be 
    changed by the randomization algorithm when using uniform seeding;
  - Correcting the design of the framelet for cluster algorithms so clusterings that are 
    not based on distance proximity to centroids are not forced to implement those methods;
  - Changing the default caching mechanism for Support Vector Machines 
    to keep rows of the kernel matrix instead of individual elements;
  - Adding methods to calculate the cache size given a number of bytes;
  - Adding cache support for Fan Chen Lin's QP (SMO's SecondOrder strategy);
  - Updating IClassifier interface to offer a NumberOfClasses property besides NumberOfOutputs;
  - Updating the base classes for IClassifier such that the Score, Probability 
    and LogLikelihood functions only have to be defined once;
  - Correcting the Score, Probability and LogLikelihood functions of GeneralizedLinearRegression;
  - Updating ITransform and IClassifier's NumberOfInputs, NumberOfOutputs and 
    NumberOfClasses properties to be read-and-write rather than read-only.

* Imaging
  - Adding the Zhang-Suen Thinning Algorithm by Hashem Zawary (thanks!);
  - Adding a FromUrl method to the Image class to download images directly from the web;
  - Adding support for jagged matrices in the ImageToMatrix and MatrixToImage converters;
  - Adding convenience methods PixelSize and Offset to UnmanagedImage;
  - Adding a constructor method in UnmanagedImage to construct from byte arrays.
  
* Vision
  - Fixing reproducibility of Bag-Of-Visual-Words when using parallel processing.

* Statistics
  - Marking Sparse kernel classes as deprecated;
  - Adding Dirac's Delta as a (non-metric) Distance function;
  - Updating the SquareEuclidean distance to support also Sparse<T> arrays;
  - Adding a dummy random number generator that generates always the same constant;
  - Fixing the implementation of the new API for Cox's Proportional Hazards;
  - Updating HMM, CRF, and HCRF learning algorithms to support creating HMMs, HMM-based classifiers, 
    CRFs and HCRFs directly from the data samples instead of requiring them to be defined by hand;
  - Adding a new MatrixContinuousDistribution base class for Wishart and Inverse-Wishart distributions;
  - Adding a RBF version of the Dynamic Time Warping kernel
    that can be used with any distance metric/cost function;
  - Updating Hidden Markov Model learning classes to use RelativeConvergence;
  - Improving discrete hidden Markov models performance;
  - Adding a base class for HCRF learning algorithms based on 
    the available IGradientOptimizationMethod optimizers;
  - Adding extension methods to simplify how distributions can be estimated 
    from the data (without requiring the distribution to be created first);
  - Fixing gradient computation in CRF learning;
  - Updating base classes for probability distributions to perform input 
    validation before calling distribution-specific implementations;
  - Adding automatic testing of all univariate probability distributions;
  - Mass fixing issues detected by automatic testing in multiple distributions.
  
* Math
  - Fixing an issue with Bounded L-BFGS in which the optimization algorithm 
    would not respect the maximum number of iterations determined by the user;
  - Adding HasConverged, MaxIterations and CurrentIteration properties 
    as required members in IConvergenceLearning interface;
  - Adding parsing methods for Vector class;
  - Fixing Sparse SquareEuclidean distance;
  - Removing duplicated extension method To<T>() from Matrix.Conversions.cs
  - Marking Conjugate Gradient and BFGS as supporting execution cancellation;

* Controls
  - Updating ImageBox to support the fluent syntax and .Hold() methods.




Accord.NET Framework 3.5.0 release notes
-----------------------------------------
21.05.2017.

Version updates and fixes:
  - GH-55: Adding support for computing TF-IDF vector representations;
  - GH-213: Linear SVM with SGD Training Support;
  - GH-297: Looking for "multinomial Logistic Regression (cross-entropy loss)" in accord like sklearn;
  - GH-330: Liblinear (Linear SVMs) does not train, exits with "index out of range";
  - GH-352: Take so much time create a SequentialMinimalOptimizationRegression in Accord.NET;
  - GH-355: SVM non-sequential 0 to n class outputs causes index out of bounds;
  - GH-365: Updating LBP to work with BoW;
  - GH-373: FrameRate as double in Accord.Video.FFMPEG;
  - GH-379: Updating NuGet specification files so assemblies are placed under net452 and net461;
  - GH-386: KNearestNeighbors Constructor (Int32, Int32, Double[][], Int32[], IMetric(Double[]));
  - GH-389: Wilcoxon Signed Rank Test | Mann-Whitney-Wilcoxon Test - differences to R;
  - GH-390: MachineLearning.KMeans: Balanced clustering;
  - GH-396: Using Parallel.For with LinearCoordinateDescent;
  - GH-397: Add an Example for OneclassSupportVectorLearning(TModel, TKernel, TInput) Class;
  - GH-398: PrincipalComponentAnalysis serialization error in v3.4.0;
  - GH-391: Add an Example for DecisionTree Class;
  - GH-392: Add an Example for GrayLevelCooccurrenceMatrix Constructor;
  - GH-393: Add an Example for GrayLevelCooccurrenceMatrix Constructor (Int32, CooccurrenceDegree, Boolean, Boolean);
  - GH-396: Using Parallel.For with LinearCoordinateDescent;
  - GH-399: Add an Example for HaarObjectDetector Class;
  - GH-401: KNearestNeighbors parallel;
  - GH-402: Add an Example for BagOfWords Class;
  - GH-405: Add an Example for RandomForest Class;
  - GH-409: Add an Example for Serializer.Load(T) Method (String);
  - GH-419: Add an Example for Combinatorics.Subsets(T) Method (ISet(T), Int32, Boolean);
  - GH-431: AugmentedLagrangian fails on standard convex quadratic problem;
  - GH-434: Possible issue with PolynomialLeastSquares() Class;
  - GH-438: Error on LocalBinaryPattern Clone();
  - GH-446: Add an Example for PolynomialRegression Class;
  - GH-448: Could not load type 'SharpDX.Bool' from assembly 'SharpDX, Version=3.1.1.0;
  - GH-451: BalancedKMeans does not find a solution for this case;
  - GH-538: CSVReader Ignores specified delimiter;
  - GH-556: Display license on repository top;
  
* General
  - Fixing target framework versions: projects that were targetting 4.5.2 and 
    4.6.1 have been updated to target 4.5 and 4.6, respectively;
  - Adding a Accord.DataSets namespace to contain classes that can download and
    pre-process well-known machine learning datasets directly from the web;
  - Adding a Accord.Text namespace to contain classes related to text processing;
  - Creating a separate Accord.Video.FFMPEG NuGet package to target x64 (Win64);
  - Upgrading C++ projects to VS2017 and removing dependencies on VS2013 and VS2015;

* Core
  - Updating the framework random generator to allow fixing the seed of independent threads:
    new random generators created from existing threads will be initialized with the global
    random seed, except if their thread-specific random seed has been manually overwritten;
  - Adding support for using GZip compression when serializing through Accord.Serializer;
  - Adding initial support for rational numbers;

* IO
  - Adding support for reading and writing NumPy's .npy and .npz formats;
  - Adding a SparseWriter to save Sparse<T> in LibSVM sparse file format;
  
* Video
  - Updating FFMPEG library to version 3.2.2;
  - Updating Accord.Video.FFMPEG to support both x86 and x64 platforms;
  - Updating FrameRate properties of VideoFileWriter and VideoFileReader 
    to be represented in Rationals instead of Int32;
  
* DataSets
  - Adding downloaders for the RCV1v2, MNIST and News20 datasets;
  
* Text  
  - Adding C# versions of Snowball's stemmers;
  
* Math
  - Fixing ToString() method of Sparse<T> vectors such that they are compatible with LibSVM;
  - Adding Angular distance metric (proper distance metric based on the Cosine similarity);
  - Updating distance metrics to be structs rather than classes, allowing for optimizations
    when used in conjunction with generic classes and methods;
  - Adding generic versions of the IOptimizationMethod interfaces, and adding a specific 
    IFunctionOptimizationMethod interface to specify function optimization methods that do 
    not necessarily require a gradient function;
  
* Statistics
  - Fixing serialization of PCA, KPCA, LDA, KDA and other classes containing a CancellationToken;
  - Adding support for learning multinomial logistic regression with any optimization algorithm;
  - Correcting/improving Wilcoxon distribution: Adding a separate method for the ComplementaryDistributionFunction
    for more precision; adding the approximation distribution using an actual NormalDistribution, and new parameters 
    at its constructor to control whether exact or approximate distribution should be used; adding parameters to 
    determine whether to apply different types of continuity corrections; adding an option to determine whether to 
    correct for ties when computing the rankings;
  - Correcting/improving Mann-Whitney U's distribution: Adding most corrections as mentioned above, plus checks to ensure
    we are computing the distribution for the case when the first sample is smaller than the second; adding a correction 
    to the variance of the approximation distribution for the case of ties; adding parallelization support when computing 
    the table of exact values;
  - Correcting/improving Wilcoxon and Mann-Whitney tests: adding a correction for ties in the variance of the Normal 
    approximation; adding checks to make sure we don't generate p-values higher than 1; adding more comparison tests 
    against results generated by R;
  - Wavelet kernel is now a struct and has support for Int32 inputs;

* MachineLearning
  - Improving the performance for LinearDualCoordinateDescent and OneVsRest SVMs;
  - Adding a ToMulticlass method to Multilabel classifiers based on scores;
  - Updating K-Nearest Neighbor to use the new classification/learning interfaces;
  - Adding parallelism to the non-data-structure-based version of K-Nearest Neighbors;
  
* Accord.Imaging
  - Adding support for LocalBinaryPattern and Haralick in Bag-of-Visual-Words.




Accord.NET Framework 3.4.0 release notes
-----------------------------------------
14.01.2017.

Version updates and fixes:
  - GH-19: Implement Grubbs' test;
  - GH-129: Possible error in Special.BSpline function;
  - GH-153: Visual Studio 2015;
  - GH-172: Add Random Forest Implementation;
  - GH-177: AugmentedLagrangian with NonlinearConstraints - Gradient NullReferenceException issue;
  - GH-183: Severity Check in NumberOfVertices Set Property on DiscreteCurveEvolution Class;
  - GH-229: Can't build cloned repository;
  - GH-250: Prediction interval - Accord.Statistics.Models.Regression.LogisticRegression;
  - GH-264: Integer division instead of double in GetSpectralResolution;
  - GH-264: Incorrect use of loop variables in sample converter;
  - GH-264: Checking same arguments multiple times in blob counter;
  - GH-264: Checking length of same vector in a loop;
  - GH-264: Integer division instead of double in Math.Tools;
  - GH-264: Dependency classes of Denavit Harternberg IK solver should be marked as Serializable;
  - GH-264: Error when checking whether component mixtures implement IFormattable;
  - GH-264: Multivariate Empirical Distribution outdated/unecessary argument checks;
  - GH-264: Correcting the support for weighted samples in Inverse Gaussian Distribution;
  - GH-275: Examples for the GoldfarbIdnani solver are not up to date and do not compile;
  - GH-291: Accord.Imaging nuget dependencies;
  - GH-295: Accord.Video.FFMPEG.VideoFileWriter ignores bitrate;
  - GH-296: Update documentation for hidden Markov models;
  - GH-299: Update to .NET 4.6 and VS2015;
  - GH-302: Regression (SVMs) : NullReferenceException on clicking 'Create Machine';
  - GH-309: Compile error with release 3.2.2;
  - GH-310: Examples for L1-regularized (Logistic) regression;
  - GH-313: Inaccuracy in Accord.Math Pseudoinverse;
  - GH-314: V3.3.0 Cannot set input and output names in LogisticRegressionAnalysis;
  - GH-320: Shared Covariance Matrix for Gaussian Mixture Models;
  - GH-325: ClusterCollection doesn't implement IEnumerable properly (runtime error);
  - GH-327: NegativeBinomialDistribution Cum Dist func not returning expected value;
  - GH-301: Bug in Accord.Statistics.Analysis.DistributionAnalysis
  - GH-304: Bug in GammaDistribution.ProbabilityDensityFunction
  - GH-330: Liblinear (Linear SVMs) does not train, exits with "index out of range";
  - GH-331: RandomForest is not serializable;
  - GH-332: Partial Least Squares issue with NIPALS method and the new API;
  - GH-337: ExpectationMaximization max Iterations can't be changed;
  - GH-340: PoissonDistribution InverseDistributionFunction not returning expected value;
  - GH-365: Can HOG to work with BoW'2 with SVM or OSVM.

* General
  - Fixing make install on Linux/Mono.
   
* Imaging
  - Updating BagOfVisualWords to implement the updated IBagOfWords interface;
  - Adding methods to facilitate the creation of BoVW with arbitrary extractors;
  - Adding examples in the documentation on how to learn SVMs on the extracted Bo(V)Ws;
  - Updating IFeatureDetector interfaces to use covariance and contravariance to avoid element-by-element type conversions.

* Math
  - Adding support for computing the full QR decomposition (besides only the economy one);
  - Adding methods to compute the null-space of a given matrix.

* MachineLearning
  - Updating the IBagOfWords interface and implementing classes to implement the IUnsupervisedLearning and ITransform interfaces;
  - Updating ZeroOneLoss to handle class labels in the -1/+1 format;
  - Updating the kernel cache to pre-compute the entire kernel matrix by default.

* Statistics
  - Adding random generators for the von-Mises Fisher distribution;
  - Updating documentation examples for Hidden Markov Models, Hidden Markov Classifiers and their respective algorithms;
  - Adding a new GammaOptions class to pass fitting options to Gamma distributions;
  - Updating DistributionAnalysis to use the new machine learning interfaces/API;
  - Updating code and documentation for Dynamic Time warping kernel;
  - Updating Gamma distribution so probabilities are computed in the log-domain by default;
  - Marking Moving and Running statistics as ISerializable;
  - Adding methods to compute the marginals in multivariate discrete distributions;
  - Adding RunningRangeStatistics and MovingRangeStatistics.




Accord.NET Framework 3.3.0 release notes
-----------------------------------------
17.09.2016.

Version updates and fixes:
  - GC-62:  Add support for computing prediction intervals in linear and generalized linear models
  - GH-113: System.AggregateException thrown in C45Learning.Run;
  - GH-115: Add documentation about how to work with sparse data;
  - GH-130: Multi class support vector machine doesn't work with SparseGaussian kernel;
  - GH-139: Examples using explicit kernel matrices;
  - GH-178: DecisionTreeWriter uses local CultureInfo when writing values;
  - GH-249: Potential bug in RandomForest or C45Learning;
  - GH-201: Adding Generalized Pareto Distribution;
  - GH-245: Incorrect method usage in Distance.GetDistance;
  - GH-255: Adding examples on how to use Laplace rule in Naive Bayes learning;
  - GH-253: BlobCounter needs a IDisposable implementation;
  - GH-252: Bug in Kurtosis Contrast Function;
  - GH-270: Adding example to show to use continuous variables in C4.5;
  - GH-271: OneclassSupportVectorLearning does not use shrinking heuristics property;
  - GH-281: Possible bug in GammaDistribution generation function when k < 1;
  - GH-282: Issue in LogisticRegression.Transform() returns true for all inputs;
  - GH-280: Merge pull request #280 from fch-aa/Fix-SMO-CacheSize;
  - GH-278: Merge pull request #278 from kulov/development;
  - GH-272: Merge pull request #272 from kdbanman/GH-271;
  - GH-269: Merge pull request #269 from mikhail-barg/minor-fix;
  - GH-273: VideoFileWriter not working;
  - GH-274: Merge pull request #274 from hzawary:development;
  - GH-285: Deserialize of Codification error in 3.2.0;
  - GH-286: Ransac - possible bug in calculation of 'N' if pInlier = 0;
  - GH-288: NaiveBayes issue when probability is 0;
  - GH-289: Incorrect use of GetLength(0) for jagged arrays in Matrix class.

* General
  - This will be last release that includes an executable installer. If you
    are still using the installer, please move to NuGet or use the framework
    compressed archive files.

* Imaging
  - Creating a new Accord.Imaging.Noncommercial assembly to hold non-commercial imaging methods;
  - Adding Fast Guided Filter to Accord.Imaging.Noncommercial.

* MachineLearning
  - Fixing Binary Split's learn method to accept null weights;
  - Updating Binary Split example to reflect the new API;
  - Adding constructors to allow tree inducing algorithms to create a tree from scratch;
  
* Statistics  
  - Fixing multiple issues with statistical analyses classes when they are used using the
    new classification/regression APIs;
  - Statistical measures (Measures.cs) have been moved to the Accord.Math assembly,
    but have been kept under the Accord.Statistics namespace;
  - Correcting L2-regularization in Logistic Regression.




Accord.NET Framework 3.2.0 release notes
-----------------------------------------
20.08.2016.

 Accord.NET 3.2 "auto-generated" release

Version updates and fixes:
  - GH-76/GC-24: Add easier creating and handling of factors for categorical variables
  - GH-123: Bug in the Euclidean on Accord.Math.Distance
  - GH-124: Fixing the Envelop filter as missing loop variables were not being incremented
  - GH-135: When the from and to ranges are equal, scaled values should remain unchanged
  - GH-159: Gamma Distribution Fit stalls for some arrays
  - GH-162: ntdll on OS X
  - GH-167: Posterior method has wrong signature in continuous hidden Markov Models
  - GH-171: Quadratic Programming (Goldfarb-Idnani) NoPossibleSolution on possible problems
  - GH-188: ProbabilisticOutputCalibration Class Example Incorrect Object Name
  - GH-206: Chessboard distance is incorrect
  - GH-214: Bug found in ReplaceChannel filter
  - GH-215: Bug Found in DecisionTrees.Learning.ID3Learning.
  - GH-225: Independent Component Analysis not converging
  - GH-232: Bug in Levenshtein distance.
  - GH-234: The subset of observations corresponding to a decision node may contains duplicates
  - GH-235: The getMaxChild method returns the max grandchild
  - GH-236: Possibly-biased comparison between errors
  - GH-237: The subset of observations corresponding to a decision node may contains duplicates
  - GH-240: Re() and Im() function of ComplexMatrix generates a OutOfRangeException

* General
  - In this release, the Matrix library from Accord.Math has been almost completely
    redesigned to make heavy use of automatic code generation. This results in more
    code reuse, more consistent interfaces and the availability of many methods which
    before were available only for Double to almost all native numerical types in the
    .NET Framework;
  - The framework now contains core classes and interfaces for defining classification
    and regression models and their respective learning algorithms, offering a more 
    standard interface when using different parts of the framework;
  - The framework now offers a Accord.Serializer class that should be responsible for
    serializing and deserializing any object from the framework, and will take care of
    versioning in case of breaking changes between releases;
  - All AForge.NET namespaces have been finally moved to inside Accord.NET, although
    some functionality is still duplicate.
  
* Core
  - Adding Interlocked operations (Increment, Add) for double values;
  - To<> universal converter can now convert jagged arrays;
  - Adding a common framework to unify all classification models, and all learning algorithms;
  - Integrating the AForge.NET Range classes in the framework, adding ByteRange;
  - Adding a common serialization mechanism to the framework to manage backwards compatibility;
  - All classes from Accord.MachineLearning.Structures have been moved into Accord.Collections;
  - Updating RedBlackTrees to implement the new base classes for tree structures;
  - Updating KD-Trees to implement the base classes for tree structures (introduces breaking changes).

* Sample applications
  - Fixing wrong arguments in sample applications.

* Math
  - Revamped matrix library making heavy use of code generation with T4 templates;
  - Matrix dot products, and elementwise operations are now auto-generated;
  - Renaming InnerProduct to Dot, and marking previous products as obsolete;
  - Vector Range, Scale and Interval are now auto-generated;
  - Standardizing the way Vectors, Matrices and Jagged matrices are created and handled
    in the framework;
  - Adding OneHot and KHot methods overloads for creating vectors using boolean masks;
  - Adding ArgMin and ArgMax methods to Vector, Jagged and Multidimensional matrices;
  - Re-implementing Matrix.Sum and Matrix.Product using T4 templates;
  - Breaking change: Sum() now computes the Sum over the entire matrix (before it needed
    to be done with Sum().Sum(). In order to compute the sum vector over rows, use matrix.Sum(0)
    and for columns, matrix.Sum(1);
  - Chessboard distance has been removed as it is the same as Chebyshev;
  - Moving AForge.NET's old Random classes into the framework, and marking them as deprecated;
  - Adding a log1pexp method for computing (1.0 + Math.Exp(-sum)) without loss of precision;
  - Adding new random generators based on Marsaglia's Ziggurat method;
  - Introducing a new, generic IRandomNumberGenerator interface so existing statistical
    distributions can be used as Random Number Generators;
  - Updating Matrix.IsEqual method to use the auto-generated overloads if possible;
  - Replacing the previous framework-wide generator with a better API;
  - Improving the framework-wide random number generator so generators created in short
    timespans do not get initialized with the same seed: Now, updating a seed will not
    affect existent random generators in other threads. It will affect only newly created
    generators and the one in the current thread;
  - Fixing the DiagonalMatrix property in SingularValueDecomposition and 
    JaggedSingularValueDecomposition so the returned diagonal matrices has the necessary 
    dimensions to reconstruct the original matrix using the decomposition main formulation;
  - Fixing a bug in Combinatorics.Sequences method where the current vector would be returned
    instead of a copy when inPlace = false;
  - Distance functions can now be auto-generated from classes from the framework;
  - Adding Dice, Jaccard, Kulczynski, Matching, Rogers-Tanimoto, Russel-Rao, Sokal Michener,
    Sokal Sneath, Yule, Bhattacharyya and LogLikelihood distances as proper classes;
  - Updating IsEqual to support absoluete and relative tolarance thresholds;
  - Adding a Histogram method for creating a histogram from an array of integer values;
  - Updating the Interval, Range and Scale method overloads to be automatically generated;
  - Adding loss functions to be used in the unified framework;
  - Moving the Elementwise class to a separate Accord.Math.Core project in order to avoid
    excessive build times due the number of auto-generated methods in this class;
  - Adding overloads to Eigenvalue decomposition to automatically sorter eigenvectors and 
    eigenvalues in descending order of absolute eigenvalue;
  - Adding a dedicated Sort static class with ordering-related methods such as Partition,
    Introsort and NthElement.
  - Expanding decompositions with two additional methods: GetInformationMatrix and Reverse
    GetInformationMatrix can be used to retrieve the standard errors for each coefficient 
	when solving a linear system; Reverse reconstructs the original matrix using the definition
	of the decomposition;
  - Deprecating Submatrix in favor of Get (methods with non-inclusive last indices);
  - Adding ArgSort function for retrieving the indices that can be used to sort a vector;
  - Adding LogSumExp to the set of special functions.

* MachineLearning
  - Adding a base foundation to encompass all classification and regression models in the
    framework as well as their learning algorithms: common interfaces and base classes for
    classifiers, distance-based classifiers and generative classifiers; common interfaces
    and base classes for supervised and unsupervised learning algorithms;
  - Updating Support Vector Machines, Decision Trees, Naive Bayes, Regressions and Analyses
    to use the new classes;
  - Unifying Linear and Kernel SupportVectorMachines, updating their classes to accept the
    Kernel function as a generic parameter: when the kernel function is a ValueType, this
    forces generic classes to be compiled specifically for each kernel type, allowing for
    the inlining of the kernel function calls;
  - Updating the way compact SVMs are represented: instead of having only a weight vector 
    and no support vectors, compact machines have a single support vector and a single weight
    of value one, eliminating what before was a special case;
  - Adding classes for OneVsOne and OneVsRest classifiers, separating the functionality that
    was previously inside MulticlassSupportVectorMachine and MultilabelSupportVectorMachine;
  - Fixing multiple issues with ErrorBasedPruning (YaronK);
  - Updating GridSearch to implement ToString methods for easier debugging;
  - Updating Linear machines and learning algorithms to accept sparse kernels;
  - Deprecating the previous sparse vector implementations and moving the current implementation
    to the existing Linear class, since they represent the same operation;
  - Adding a true implementation for LibSVM-style Sparse vectors;
  - Updating SparseReader to read sparse vectors using the new Sparse representation;
  - Refactoring the clustering namespace to increase code reuse between the different algorithms;
  - Updating K-Means, GMM and BagOfWords to expose a ParallelOptions object that can
    be used to configure and stop the parallelization of those algorithms;
  - Updating K-Means to support sample weights;
  - Correcting multiple random initializations of Gaussian mixture model;
  - Adding a PriorityQueue class based on the MIT-licensed code by Daniel "BlueRaja" Pflughoeft.
  - Adding Vantage-Point and Space-Partitioning trees and Barnes Hutt t-SNE based on the original
    code from Laurens van der Maaten BH t-SNE implementation;
  - Adding a basic implementation for the Apriori algorithm.

* Imaging
  - Updating static methods in AForge.NET's Image class to become extension methods;
  - Implementing ICloneable in all corner and feature detectors.

* Neuro
  - Updating ResilientBackpropagation with the improvements from iRProp+.

* Statistics
  - Adding Non negative Least Squares regression;
  - Adding Procrustes Analysis;
  - Deprecating IAnalysis in favor of the new framelet for classification, 
    regression and transformation methods;
  - Merging AForge.NET and Accord.NET Histogram classes;
  - Updating IFittingOptions to implement ICloneable;
  - Adding constructors to Independent distributions accepting a lambda function
    to initialize inner components instead of relying on cloning;
  - Adding a Classes class to provide methods that operate with categorical/label data,
    such as converting boolean, double or integer values to [0;1] or [-1; +1] indicators;
  - Adding Decide methods to unambiguously transform a distance/score value into a boolean;
  - Updating statistic distributions to implement the IRandomNumberGenerator interface, meaning 
    any distribution can now be used as random number generator;
  - Adding the Metropolis-Hasting sampler to generate samples from multivariate distributions 
    that do not have specialized samplers;
  - Adding named constructors for building regressions directly from coefficient vectors;
  - Updating kernels to rely in Accord.Math.IDistance instead of the previous IDistance from 
    the Statistics namespace;
  - Adding Pearson's Universal Kernel, Thin Spline Plate and Hellinger kernels
    contributed by Diego Catalano;
  - Moving standard statistical measures (i.e. mean, standard deviation, variance, ...) to a
    separate Measures class;
  - Updating Mean methods to operate in the same way as Sum: if a dimension is not specified,
    the Mean will be computed across all dimensions of the matrix;
  - Updating Hidden Markov Models to use the new Tagger interfaces and base classes.

* Genetics
  - Updating the Genetics project to use the new sample generators based on statistical 
    distributions;
    

    

Accord.NET Framework 3.0.0 release notes
-----------------------------------------
16.08.2015.

Version updates and fixes:
  - GC-70: Merge with AForge.NET.
  - GC-90: Convert unit test projects to NUnit
  - GH-114: GeneralizedBetaDistribution's calls Random with wrong parameter order

* General
  - This release marks a milestone in the Accord.NET Framework. Starting from this
    release, the AForge.NET Framework has been incorporated directly in the project,
    meaning that we are now free to fix, maintain, transform and improve AForge.NET
    directly.

  - This release provides most of the AForge.NET namespaces unchanged. This means
    that this specific version of Accord.NET Framework can be used as drop-in
    replacement in any project currently using the AForge.NET Framework and that
    is willing to upgrade to Accord.NET sometime in the future.

  - This release is mostly a transition release to help projects using the AForge.NET
    framework make the transition to Accord.NET more easily. Further releases will be
    aimed at improving the interaction between the two codebases and streamlining the
    provided functionality.




Accord.NET Framework 2.15.0 release notes
-----------------------------------------
01.05.2015.

Version updates and fixes:
  - GC-56:  Reuse decision attributes in the C4.5 algorithm for decision trees;
  - GC-109: The GoldfarbIdnani optimizer does not optimize well some problems;
  - GH-24:  Disentangling unit test projects and adding a LGPL-only project;
  - GH-57:  Decision trees created using C4.5 depends on the sorting order;
  - GH-58:  SURF detector might generate Divide By Zero exceptions;
  - GH-60:  Regularization breaks LogisticRegressionAnalysis in 2.14;
  - GH-61:  SVM code that worked with version 2.11 now fails to converge;
  - GH-64:  Exception in KPCA when using jagged matrices;
  - GH-69:  Fix K-Means deserialization between framework versions.
  
* General
  - Upgrading solution to VS2013 and adding support for .NET 4.5;
  - Packaging scripts can now create NuGet symbol packages;
  - The framework can now be built using Mono.

* Accord.Statistics
  - Correcting Circular statistics' AngularDeviation method;
  - Correcting kernel profile functions and their gradient;
  - Improving the precision of the Binomial distribution;
  - Improving sample generation for Poisson and Rayleigh distributions;
  - Updating all univariate distributions to support sample generation;
  - Making sure all probability distributions implement IFormattable;
  - Adding Generalized Beta distribution with PERT estimation;
  - Adding the von-Mises Fisher distribution for circular data;
  - Adding sample generation in Beta and Generalized Beta distributions;
  - Adding estimation using the Method-of-moments and Maximum Likelihood;
  - Adding support for weighted samples in LogisticRegressionAnalysis;
  - Adding a named constructor to create an Analysis from summary data;
  - Adding all missing Shapiro-Wilk distribution's methods;
  - Adding a common interface for radial basis function kernels;
  - Adding a new generic Gaussian kernel for creating composite kernels;
  - Adding a Windowing filter in the Statistics filters namespace;
  - Adding support for weighted samples in LogisticRegressionAnalysis;
  - Adding a named constructor to create an Analysis from summary data;
  - Adding Multinomial Logistic Regression Analysis.

* Accord.Math
  - Updating Augmented Lagragian to detect more accurately 
    when the inner optimization algorithm has diverged;
  - Adding a new Fast Fourier Transform (FFT) implementation for general 
    matrices and vectors whose dimensions are not necessarily powers of 2;
  - Adding Hellinge and Levenshtein distances for generic arrays;
  - Adding jagged-matrix version of the QR and Eigenvalue decompositions.

* Accord.MachineLearning
  - Updating tree inducing algorithms (ID3 and C4.5) se they can reuse
    decision variables multiple times when creating a decision tree;
  - Correcting Levenberg-Marquardt's chain-rule Jacobian calculation 
    when there are many output neurons in the learned neural network;
  - Updating the way SVM learning algorithms detect whether a machine is linear or not;
  - Updating SVM learning algorithms to use an heuristic value for 
    C by default unless it has been manually specified by the user;
  - Updating the linear kernels so they are created without a constant term by default;
  - Improving multi-class SVM to generate more user-friendly stack traces when
    an exception occurs during the learning of one of the binary sub-problems;
  - Updating Kd-Trees to use interval heaps instead of general .NET structures;
  - Adding Nu-SVMs based on LibSVM's quadratic programming solver.

* Accord.Neuro
  - Updating Levenberg-Marquardt to avoid setting lambda to zero. 

* Accord.Imaging
  - Updating IntegralImage to work with In64 matrices to avoid overflows;
  - Correcting Variance, Sauvola and Niblack threshold filters;
  - Adding Fast Variance, Wolf-Joulion Threshold, RG Chromaticity and 
    Objective fidelity filters.

* Accord.IO
  - Integrating and repackaging Sebastien Lorion's Fast CSV Reader into the Accord.IO
    namespace with an added support for auto-detecting the file's field delimiter.

* Accord.Audio
  - Adding IWindow apply overloads to operate directly on double[] vectors;
  - Adding Sine and Custom signal generators and correcting existing ones.
  
* Sample applications
  - Adding feature selection sample application using L1-regularized logistic SVMs;
  - Correcting the display of all sample applications in high-DPI displays.




Accord.NET Framework 2.14.0 release notes
-----------------------------------------
15.12.2014.

Version updates and fixes:
  - GH-29: HiddenConditionalRandomField is not correctly serialized;
  - GH-31: Adding hidden Markov Model methods for computing the probability of 
           a state assuming a particular value inside an observation sequence;
  - GH-36: Extensions class collides with the Accord.NET Extensions Framework;
  - GH-35: Distribution issue with .NET 3.5 assemblies;
  - GH-37: Adding Taylor series functions and dissimilarity functions;
  - GH-42: Adding new contributed distance functions;
  - GH-46: Optimization functions and constraints to now support different cultures;
  - GH-48: Fix calculation of log likelihoods for BaumWelch learning algorithm;
  - GC-33: GoldfarbIdnaniQuadraticSolver class failed to give correct answer.

* Accord.Core
  - Adding Red-Black trees and a Red-Black dictionary based on those trees
    with support for efficiently searching for maximum and minimum elements.

* Accord.IO
  - Updating the IDX reader to automatically convert between different data types.

* Accord.Math
  - Adding the Nelder-Mead and Subplex non-linear optimization algorithms;
  - Adding matrix padding methods (adding extra rows and columns with zeros);
  - Updating quadratic optimization constraints to support tolerance parameters.

* Accord.Statistics
  - Adding regularization in IterativeReweightedLeastSquares;
  - Adding DistributionAnalysis for estimating distributions from observed data;
  - Adding weighted measure methods in Statistics.Tools that are 
    based on element repetitions rather than element importance;
  - Adding Anderson-Daring, Shapiro-Wilk, Inverse Chi-Square, Lvy, Folded 
    Normal, Shifted Log-Logistic, Kumaraswamy, Trapezoidal, U-quadratic and
    BetaPrime distributions;
  - Adding Anderson-Daring and Shapiro-Wilk hypothesis tests;
  - Correcting the MarkovMultivariateFunction constructor for 
    explicit Independent<NormalDistribution> hidden Markov models;
  - Correcting discrete Viterbi learning convergence check and unifying 
    the Viterbi implementations for discrete and continuous variables;
  - Correcting the calculation of log likelihoods in Baum Welch learning;
  - Correcting serialization issue with DynamicTimeWarping kernel;
  - Updating all distribution functions constructors to offer Range attributes
    that can be used to automatically determine valid values for its parameters;
  - Updating Logistic Regression and Cox's Proportional Hazards 
    analyses to avoid computing LR-ratio tests unless necessary;
  - Updating Multivariate Empirical Distributions to support CDF;
  - Updating Empirical Distributions to support weighted samples;
  - Updating CRF output feature functions to activate only once per sequence;
  - Updating all probability distributions to offer a Generate method;
  - Updating ChiSquare Test to support testing against continuous distributions;
  - Updating univariate discrete distributions with Quantile Density functions.

* Accord.MachineLearning
  - Adding LIBLINEAR's linear support vector regression (SVR) algorithms;
  - Adding LinearCoordinateDescent for the primal formulation and renaming the 
    previous coordinate descent algorithm to LinearDualCoordinateDescent;
  - Adding named constructors in Naive Bayes to build Gaussian models more easily;
  - Updating Naive Bayes classifiers to estimate component variables in parallel;
  - Updating ID3 algorithm so attributes can join multiple times a decision;
  - Updating Bag-of-Visual-Words to avoid computing costly cluster measures;
  - Updating K-Means to support setting algorithm options through its class itself;
  - Updating K-NearestNeighbor feature matching to use KD-Trees when its possible;
  - Updating error-based pruning to avoid computing modes when there are no elements;
  - Updating K-Modes to select modes per column, instead of entire elements. Now it 
    is also possible to use the Kmeans++ initialization scheme in this algorithm.
 
* Accord.Neuro
  - Adding Rectified Linear activation functions.

* Accord.Imaging
  - Fixing Niblack and Sauvola thresholding algorithms for 8bpp images.

* Accord.Audio
  - Adding a Volume adjustment filter for audio signals.

* Accord.Controls
  - Adding a DataBarBox visualization box;
  - Updating visualizations to offer more flexibility in customizing ZedGraph charts;

* Sample applications
  - Adding support for adjusting volume in the Recorder sample application.




Accord.NET Framework 2.13.1 release notes
-----------------------------------------
30.08.2014.

Version updates and fixes:
  - GC-61: GoldfarbIdnani should implement the IOptimizationMethod interface;
  - GC-63: Updating optimization methods to avoid throwing exceptions;
  - GC-80: Audio.Formats.WaveDecoder.AverageBitsPerSecond should be bytes, not bits;
  - GC-85: Tools.Random is not thread safe;
  - GC-87: Add a base distance method for kernel functions;
  - GC-88: General Discrete Distribution estimation doesn't handle negative values;
  - GC-90: EmpiricalDistribution's SmoothingRule should use +0.2 in the factor calculation;
  - GC-91: Update L-BFGS optimization algorithms from the latest Fortran source algorithms;
  - GC-96: ConvexHullDefects doesn't accounts for defects between the last and first points;
  - GC-98: Fixing ProbabilityDensityFunction in MultivariateNormalDistribution;
  - GC-99: Combinatorics.Permutations(array) method does not include the original array;
  - GC-101: Issue with Haar transform at orders grater than one;
  - GH-04: Replacing SlimDX with SharpDX in Audio libraries;
  - GH-05: ComplexSignalConstructor unit test fails;
  - GH-07: Fixing typos in Haralick algorithm;
  - GH-08: Updating IntegralImage2 to support Format32bppRgb images;
  - GH-09: Binomial probability mass function result differs from Excel result;
  - GH-10: Fixing issue in Singular value decomposition clone method;
  - GH-11: Fixing issue in Statistics.Filters.Codification.Translate method;
  - GH-12: Add a HistogramBox viewer in the same spirit as ScatterplotBox and DataGridBox;
  - GH-13: NaiveBayes constructor incorrectly checks its arguments;
  - GH-25: SauvolaThreshold and NiblackThreshold filter are always making whole black image;
  - GH-27: Fixing Bernoulli distribution sampling.

* General
  - This release brings some big changes. The SlimDX libraries have been replaced by SharpDX, 
    which now enables the framework to be run into more platforms. Two new assemblies have also 
    been created to further accommodate future framework extensions, such as a dedicate matrix
    format library, and an additional library for math algorithms that are not available under 
    the LGPL. As such, the former Accord.Statistics.Formats namespace has been moved into the
    new Accord.IO, and a new Accord.Math.NonCommercial assembly has been created to contain
    other algorithms that are free but are not shared under a free software license and thus 
    cannot be used in commercial applications.

* Accord.IO
  - Adding a LibSVM's file model to load and save LibSVM's classifiers;
  - Adding a IDX file reader to read the MNIST (and possibly others) datasets;
  - Adding a MAT file reader to read Matlab/Octave data, including sparse,
    structures and objects.
    
* Accord.Statistics
  - Fixing Codification filter when columns have a max-length constraint;
  - Fixing MultivariateNormalDistribution to handle edge cases;
  - Fixing distance computations for most kernel functions;
  - Updating Hidden Markov Model classes with more Debug assertions;
  - Updating descriptive analysis to support quantile computation;
  - Adding Deviance measure in the DescriptiveAnalysis;
  - Adding a IReverseDistance interface to separate distinct kernel distances;
  - Adding support for directly processing matrices in the Normalization filter;
  - Adding support for explicit feature space transformation in some kernels;
  - Adding Beta and Pareto distribution fitting;
  - Adding explicit Gaussian kernel projection using a Taylor approximation;
  - Adding Birnbaum-Saunders, Generalized Normal, Gumbel, Power Lognormal, 
    Power Normal, Triangular, Tukey Lambda, Logistic, Hyperbolic Secant,
    Degenerate and General Continuous distributions.

* Accord.Imaging
  - Adding new user-contributed imaging filters: Difference of 
    Gaussians, HighBoost, Niblack Threshold, Sauvola Threshold.

* Accord.Math
  - Introducing the Accord.Math.Integration namespace for numerical integration;
  - Updating FiniteDifferences to support any order and number of interpolation points;
  - Fixing NonnegativeFactorization when smaller rank approximations are requested;
  - Updating the previous L-BFGS implementation from the latest Lbfgsb.3.0 version;
  - Adding the L-BFGS optimizer with support for Orthant-Wise optimizer from LibBFGS;
  - Adding the Cobyla method for derivative-free optimization (user contribution);
  - Adding Carl Edward Rasmussen's FminCG conjugate gradient minimizer (user contribution);
  - Adding the Discrete Curve Evolution algorithm (user contribution);
  - Adding a Singular value decomposition implementation for Jagged matrices;
  - Adding a simple 2D Kalman filter implementation (user contribution);
  - Adding an 'online' version of the TruthTable method which can generate samples on the fly
    through IEnumerable. Now it can also work with any number of symbols per element position.

* Accord.Audio
  - Fixing position calculation when sampleIndex is set in WindowBase.Apply method.

* Accord.MachineLearning
  - Fixing SequentialMinimalOptimization when working with weighted samples;
  - Renaming ProbabilisticOutputLearning to ProbabilisticOutputCalibration;
  - Updating cross-validation to support class-balanced partitions;
  - Adding support for different sample weights in SMO and and IRLS;
  - Adding a configurable model threshold for AdaBoost learning;
  - Adding a generic base model for AdaBoost's weak classifiers;
  - Adding specialized linear SVM algorithms based on liblinear's offerings, 
    such as L2-regularized L1 and L2-loss SVM for the dual formulation and
    the L2-regularized, L2-loss SVM for the primal formulation, as well as
    probabilistic SVM and Logistic Regression algorithms.

* Accord.Controls
  - Adding HistogramBox for displaying histograms in the same manner as MessageBox;
  - Updating Scatterplot to allow for constructing a plot from a single parameter.



Accord.NET Framework 2.12.0 release notes
-----------------------------------------
04.01.2014.    

Version updates and fixes:
  - GC-83: Error in Accord.Statistics.Tools;
  - GC-81: Redundant condition check in specialized Inverse method;
  - GC-82: Serializable annotation is needed on class contrast functions;
  - GC-66: Adding fitting options for empirical distributions.

* General
  - The general focus for this release was again to improve 
    the documentation and provide standard bug-fixing;
  - The project now also provides Debug binaries which be used
    to provide more detailed information when debugging applications.

* Accord.Controls
  - Adding a WavechartBox control to display wavecharts with ease, in 
    the same way as MessageBox can display text messages on screen.

* Accord.Audio
  - Correcting WaveDecoder.AverageBytesPerSecond (Google Code #80);
  - Correcting the creation of base audio windows;
  - Adding the ExtractChannel filter for extracting single channels
    from multiple-channel audio signals.

* Accord.Statistics
  - Adding generic base classes for probability distributions;
  - Adding support for computing the cumulative multivariate normal 
    distribution function for specifically one and two dimensions;
  - Correcting behavior of the Binomial test under .NET 4.5;
  - Updating DescriptiveAnalysis to provide sums and confidence intervals;
  - Updating ChiSquareTest to accept ConfusionMatrices as input;
  - Unifying Univariate and Multivariate mixture distribution fitting
    through Expectation-Maximization into a single and generic class;
  - Updating the WeightedMean method to accept unnormalized weights;
  - Updating the Codification filter to work without DataTables;
  - Adding fitting options for empirical distributions (Google Code #66);
  - Adding options to robustly fit Multivariate Normal Distributions using 
    the Singular Value Decomposition, avoiding non-positive definite issues. 

* Accord.MachineLearning
  - GaussianMixtureModel now accepts a maximum number of iterations;
  - Updating the KDTree class to provide a non-generic version when there 
    is no interest in the kind of values stored as information in the nodes;
  - Adding specialized methods for when only the closest neighbor is needed
    in a KDTree. Also adding options to provide only an approximate answer;
  - Updating the K-Means algorithm to fully support parallel processing;
  - Adding support for generating decision rules from Decision Trees.

* Accord.Math
  - Correcting InsertColumn and InsertRow implementations;
  - Improving Submatrix creation to avoid extra memory allocations when desired;
  - Reshape method for jagged arrays can now work with non-rectangular arrays;
  - Correcting relative convergence issues with negative watched interest values;
  - Adding Modulo and Modular distance for double-precision floating point inputs.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.2.5. May work with newer versions.



Accord.NET Framework 2.11.0 release notes
-----------------------------------------
27.10.2013.    

Version updates and fixes:

* General
  - The general focus for this release was to improve the documentation, 
    re-organize the sample applications and provide standard bug-fixing.

* Accord.Controls
  - Fixing ArrayDataView to show jagged arrays in DataGridViews;
  - Adding a ComponentView control to display analysis information.

* Accord.Imaging
  - Adding a Save overload to BagOfVisualWords (Google Code #74);
  - Marking FAST and FREAK as Serializable (Google Code #75).

* Accord.Statistics
  - Improving documentation for the 'unbiased' parameter (Google Code #66);
  - Removing the DEBUG flag from the Statistics assembly (Google Code #78);
  - Adding NonlinearRegression models and least-squares learning algorithms;
  - Adding a option to force the use of SVD when fitting linear regressions;
  - Adding a option to disable normalization when creating Cox's models;
  - Adding MultipleBaumWelchLearning to test random weight initializations;
  - Updating HCRF learning algorithms to use relative convergence by default.

* Accord.MachineLearning
  - Correcting the score generation on k-NN classifier (Google Code #68);
  - Fixing the Compact property of the SequentialMinimalOptimization class;
  - Correcting the KD-Tree implementation to correctly identify all neighbor
    points when querying for a finite number of neighbors (Google Code #71).

* Accord.Math
  - Adding the Gauss-Newton and Levenberg-Marquardt Least-Squares algorithms;
  - Updating Absolute and Relative Convergence to implement a common interface;
  - Adding a configurable minimum convergence checks in relative convergence.

* Sample applications
  - Reorganizing all sample applications, adding more relevant information
    to their description and improving their documentation.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.2.5. May work with newer versions.




Accord.NET Framework 2.10.0 release notes
-----------------------------------------
07.09.2013.    

Version updates and fixes:

* General
  - This release aimed to provide improvements to the documentation. 
    Most of the Univariate Distributions now include proper examples
    for all main functions and measures in their summary page.

* Accord.Imaging
  - Adding Haralick's set of textural features;
  - Adding Local Binary Pattern (LBP) features;
  - Adding Gabor, Variance and Kuwahara filters;
  - Adding Gray-World, WhitePatch filters;
  - Adding Kirsch and Robinson edge detectors;
  - Adding Gray-Level Co-occurrence Matrices (GLCM);
  - Adding Gray-Level Run-length Matrices (GLRLM);
  - Adding Gray-Level Difference Method (GLDM);
  - Adding byte and color conversions in the ArrayToImage converter;
  - Updating IFeaturePoint's to include better conversion operators.

* Accord.Statistics
  - Adding random Covariance matrix generation methods;
  - Adding Von-Mises cumulative distribution function;
  - Adding Poisson inverse cumulative distribution function;
  - Adding Support information for all distributions;
  - Adding support for threshold models in Markov running filters;
  - Adding default support computing the true Median and Quantile 
    functions (inverse cumulative distribution function) for all 
    univariate continuous distributions;
  - Improving ToString methods for linear/polynomial regressions;
  - Updating all regression methods to use SVD by default;
  - Correcting Wishart and Inverse Wishart distributions.

* Accord.MachineLearning
  - Adding simple Minimum (Mean) Distance Classifier;
  - Adding RANSAC methods for plane, circle and line fitting;
  - Adding options to trade speed vs. accuracy in Mean-Shift; 
  - Updating Bag-of-Words to support any feature point type;
  - Fixing Grid-Search issue when one of the models returns NaN;
  - Fixing issue when k-Nearest Neighbors Matching is called 
    with less than k points on its second argument;
  - Improving the performance of KD-Trees and Mean Shift;
  - Updating Gaussian Mixture Model methods to report the 
    classification strength (score) as an out parameter.

* Accord.Math
  - Adding the first version of the Math.Kinematics
    namespace with Denavit-Hartenberg joint models;
  - Adding Binary Search root finding algorithm;
  - Adding missing Gamma.Inverse special function;
  - Adding Mixture Distribution's Distribution functions;
  - Adding dedicated class for Gram-Schmidt Orthogonalization;
  - Adding general purpose Resilient Backpropagation algorithm; 
  - Fixing precision-loss issue in the Augmented Lagrangian Solver;
  - Updating FiniteDifferences to support configurable step sizes;
  - Updating Matrix.Range methods to require a dimension parameter;
  - Moving matrix formatters from Accord.Math.Formats to 
    Accord.Math in order to improve their discoverability.

* Accord.Vision
  - Moving the GroupMatching algorithms to Accord.Vision.

* Sample applications
  - Adding sample application for Denavit-Hartenberg joint models;
  - Adding sample application for Eigenfaces, PCA for images;
  - Adding sample application for mouse gesture classification 
    using Dynamic Time Warping Kernel Support Vector Machines. 
  
* AForge Compatibility
  - Compiled against AForge.NET Framework 2.2.5. May work with newer versions.




Accord.NET Framework 2.9.0 release notes
-----------------------------------------
20.04.2013.    

Version updates and fixes:

* General
  - Updating installer to register framework libraries.

* Accord.MachineLearning
  - Adding Least-Squares Support Vector Machines (LS-SVMs);
  - Adding support for Decision Tree pruning (experimental);
  - Adding Hybrid Markov Models (experimental, use with care);
  - Adding Load and Save methods to Bag-of-Words;
  - Adding initial support for Boosting (AdaBoost);
  - Correcting k-Nearest Neighbors, KD-Trees and 
    Mean-Shift to respect non-Euclidean distances;
  - Fixing IndexOutOfRange Exception in C4.5 learning;
  - Fixing bug in k-Nearest Neighbors classifier;
  - Fixing C4.5 split when working with continuous attributes;
  - Renaming DecisionAttributeKind to DecisionVariableKind
    for better naming consistency with DecisionVariables.

* Accord.Statistics
  - Adding support for combining Confusion Matrices;
  - Updating Levene's Test to support Trimmed Mean;
  - Updating Binomial Test to use the small p-values;
  - Fixing Two-Tailed computation in Binomial test.

* Accord.Math
  - Adding Discrete Cosine, Sine and Hartley Transforms.

* Accord.Imaging
  - Adding missing overloads for image moment calculators;
  - Adding the Fast Retina Keypoint (FREAK) image descriptor;
  - Correcting Histograms-of-Oriented-Gradients;
  - Updating HaarObjectDetector to ignore Channel 
    information when working with Grayscale images.

* Accord.DirectSound
  - Adding support for recording multiple bit-rates.

* Accord.Controls
  - Adding support for Codebooks in DecisionTreeView.
  
* AForge Compatibility
  - Compiled against AForge.NET Framework 2.2.4. May work with newer versions.





Accord.NET Framework 2.8.2 release notes
-----------------------------------------
25.02.2013.    

Version updates and fixes:

* Accord.MachineLearning
  - Adding several K-d tree traversal methods;
  - Adding support for generating decision tree assemblies;
  - Adding maximum height option for ID3 and C4.5 learning;
  - Adding a generic Bag-Of-Words to accept other feature detectors;
  - Updating DecisionVariable to initialize itself from a Codebook;
  - Fixing C4.5 learning when presented with constant variable inputs;
  - Fixing and normalizing the name of cluster identification methods.
  
* Accord.Statistics
  - Adding support for opening binary streams in ExcelReader;
  - Adding support for generalized linear regression models;
  - Adding support for integer observations in generic density HMMs;
  - Fixing Levene's test to behave as a one-tailed test;
  - Fixing Stepwise Logistic Regression's result property;
  - Fixing HMM classifier rejection using Threshold models;
  - Fixing ConfusionMatrix's ToGeneralMatrix method so it respects
    the expected order of elements in the contingency table.

* Accord.Math
  - Adding Discrete Cosine, Sine and Hartley Transforms.

* Accord.Imaging
  - Adding Hu's set of invariant image moments;
  - Adding k-nearest neighbors feature point matching;
  - Adding an early, experimental version of the Histograms of
    Oriented Gradients (HOG) descriptors due to popular demand;
  - Correcting image converters for different pixel sizes.

* Accord.DirectSound
  - Adding support for capturing audio in different bit resolutions.
  
* Accord.Controls
  - Fixing the initialization of ScatterplotView.
  
* AForge Compatibility
  - Compiled against AForge.NET Framework 2.2.4. May work with newer versions.




Accord.NET Framework 2.8.1 release notes
-----------------------------------------
21.12.2012.    

Version updates and fixes:

* General
  - Working on more source code examples for the documentation.

* Accord.Math
  - Adding Levenshtein distance for strings.

* Accord.Imaging
  - Updating BagOfVisualWords to be fully serializable.

* Accord.MachineLearning
  - Adding K-dimensional trees (K-d trees);
  - Adding Mean-Shift clustering algorithm;
  - Adding support for weights in Gaussian Mixture Models;
  - Correcting the name of the K-Nearest Neighbors algorithm;
  - Improving K-Nearest Neighbors for double[] using a K-d tree;
  - Changing K-Nearest Neighbors generic argument to represent the 
    instance type rather than the type of the array of instances.

* Accord.Statistics
  - Adding interfaces for density estimation kernels;
  - Adding Gaussian, Epanechnikov, Uniform density kernels;
  - Adding Bartlett's and Levene's tests for variances;
  - Adding hypothesis tests for comparing ROC curves;
  - Adding support for scatter plot generation directly from ROC curves;
  - Adding running Markov models and running Markov classifier filters;
  - Adding stochastic gradient descent learning for logistic regression models;
  - Updating Two-Sample Kolmogorov to work with number of samples in double-precision;
  - Correcting standard error calculation in Paired T-Tests.

* Accord.Neuro
  - Adding Boltzmann Machines, Contrastive Divergence Learning and Deep Neural Networks.

* Accord.Vision
  - Adding rectangle averaging support in the Haar object detector.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.2.4. May work with newer versions.




Accord.NET Framework 2.8.0 release notes
-----------------------------------------
06.11.2012.    

Version updates and fixes:

* General
  - Adding support for previous versions of the .NET framework (3.5).

* Accord.Audio
  - Fixing finalizers to avoid disposing null object references.

* Accord.Math
  - Adding Absolute and Relative convergence classes;
  - Adding Algorithm Environments to the Math project;
  - Fixing SVD precision when solving ill-conditioned problems.

* Accord.Statistics
  - Reorganizing the (Hidden) Conditional Random Fields namespace;
  - Adding Forward-Backward Gradient calculator for linear-chain HCRFs;
  - Adding Cox's Proportional Hazards and Partial Newton-Raphson estimation algorithm;
  - Adding Gompertz and Empirical Hazard Distributions;
  - Adding Weighted Confusion Matrix and Weighted Kappa;
  - Adding support for Weighted Kappa in Kappa tests;
  - Fixing class balancing in the Grouping filter;

* Accord.MachineLearning
  - Marking DecisionAttributeCollection as Serializable;
  - Updating Multi-class and Multi-label SVMs to implement the IEnumerable interface;
  - Updating Multi-class SVM decision to be thread-safe in multithreaded environments;
  - Adding explicit class count parameter to KNearestNeighbors;
  - Adding Bag-of-Words model for text categorization;
  - Fixing SMO when there is only one class label in outputs;
  - Fixing set size counting in Cross-Validation.

* Accord.Controls
  - Updating Confusion Matrix View to display Actual and Expected labels.

* Accord.Imaging
  - Fixing an issue with PointH transform method.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.2.4. May work with newer versions.




Accord.NET Framework 2.7.1 release notes
-----------------------------------------
01.09.2012.    

Version updates and fixes:

* Accord.Statistics
  - Adding data processing filters for data randomization,
    branching, grouping, elimination and imputation;
  - Adding Precision, Recall and F-Score in Confusion matrix; 
  - Adding GrandMean calculation in Statistics.Tools;
  - Adding the Negative Binomial, Pareto, Laplace, Inverse
    Gamma, Beta, Wishart, Inverse Wishart, Dirichlet distributions;
  - Adding the Two Proportion Z-Test, Paired T-Test and Multinomial test
  - Correcting Noncentral T distribution pdf calculation; 
  - Correcting Skewness and Kurtosis calculations; 
  - Correcting ANOVA when groups have different sizes;
  - Adding support for newer Excel versions in ExcelReader;
  - Adding support for different sample weights in HMM learning;
  - Adding distinct elements calculation in DescriptiveAnalysis;
  - Adding standalone Standard Error computation for ROC curves. 

* Accord.MachineLearning
  - Redesigning clustering framework to accommodate the
    Binary-Split algorithm, K-Means, K-Modes and GMMs;
  - Adding support for cancellation in SVM learning;
  - Adding support for different class weights in SMO;

* Accord.Controls
  - Adding Confusion Matrix viewer;
  - Adding numeric collection editor.

* Accord.Imaging
  - Adding Mean and Standard Deviation for images.

* Accord.Vision
  - Marking TrackingObject as Serializable.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.2.4. May work with newer versions.




Accord.NET Framework 2.7.0 release notes
-----------------------------------------
08.07.2012.    

Version updates and fixes:

* Accord.Statistics
  - Standardizing statistical tests hypotheses;
  - Adding power analysis for statistical tests;
  - Adding Fisher's exact test for contingency tables;
  - Adding Multiple-Matrix Kappa and Average Kappa tests;
  - Adding Binomial tests, Sign tests, Wilcoxon and Mann-Whitney tests;
  - Adding Noncentral T, Wilcoxon's W and Mann-Whitney's U distributions;
  - Adding Multiple Linear Regression Analysis with ANOVA statistics;
  - Adding inverse distribution calculation in the T-Distribution;
  - Adding the unsupervised Viterbi learning and the supervised
    Maximum Likelihood Estimation learning for hidden Markov models;
  - Adding Generate methods for sampling from hidden Markov models;
  - Marking several Statistics.Tools methods as extension methods;
  - Introducing the ISampleableDistribution interface for probability
    distributions and adding generate methods for most distributions;
  - Correcting variance calculation for Cohen's Kappa, adding variants
    for the null hypothesis and the calculation using the Delta method;
  - Correcting F-Distribution ComplementaryDistributionFunction;
  - Correcting Kernel Principal Component Analysis bug when auto-
    selecting the maximum number of components;
  - Resolving naming confusion in KMeans, renaming Classify to Nearest;
  - Renaming ContinuousUniformDistribution to UniformContinuousDistribution;
  - Renaming DiscreteUniformDistribution to UniformDiscreteDistribution;
  - Renaming Random to RandomSample for clarity;
  - Updating MultipleLinearRegression to use a single SVD;
  - Updating Covariance method with two variables to generate
    a covariance value rather than a 2x2 covariance matrix;
  - Renaming IMultipleRegressionAnalysis interface to the
    more adequate name IMultivariateRegressionAnalysis;
  - Renaming ConfusionMatrix.Observations to .Samples to  
    maintain consistency with the GeneralConfusionMatrix.

* Accord.MachineLearning
  - Adding Bootstrap for generalization performance estimation;
  - Adding the K-Modes clustering algorithm for discrete data;
  - Optimizing both memory usage and performance for Sequential
    Minimal Optimization learning and the kernel function cache.

* Accord.Math
  - Adding Brent Search Root Finding Algorithm;
  - Adding Combinatorics class for permutations and combinations;
  - Fixing a bug in Goldfarb-Idnani Quadratic Solver;
  - Adding Trigamma function to Gamma class;
  - Adding array format providers.

* Accord.Controls
  - Adding ScatterplotBox class for displaying point data; DataGridBox for
    displaying tabular data; and DataSeriesBox to display series data in the
    same way as Windows.Forms's MessageBox can be used to display text data;
  - Adding extension method for enabling DataGridViewColumns nested properties;
  - Adding ImageBox overloads to work with images represented in matrix form.

* Accord.Imaging
  - Adding the Bag of Visual Words (BoW) feature extractor;
  - Renaming SpeededUpRobustFeatures (SURF) classes avoiding acronyms.
  - Updating FeaturesMarker to make it work with indexed images.

* Sample applications
  - Adding image classification sample application;
  - Updating Regression sample application to include
    more information for linear regression analysis.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.2.4. May work with newer versions.




Accord.NET Framework 2.6.1 release notes
-----------------------------------------
05.05.2012.    

Version updates and fixes:

* General
  - The project's website has moved. This minor release updates all
    references to reflect the new project URL and includes only some new
    additions and fixes. The project is now being hosted at Google Code.

* Accord.MachineLearning
  - Adding Multi-label Support Vector Machines;
  - Adding K-Nearest Neighbor (k-NN) classifiers;
  - Adding linear elimination Support Vector Reduction;
  - Correcting SVM serialization compatibility.

* Accord.Statistics
  - Adding named constructors in PCA to create analysis
    directly from covariance or correlation matrices;
  - Adding support for adjusting the rejection threshold
    in Hidden Markov Classifiers with threshold models;
  - Updating Center, Standardize and ZScores methods so
    they are not in-place by default.

* Accord.Math
  - Adding Augmented Lagrangian solver for Non-linear programming problems;
  - Adding Row-reduced Echelon Form reductor.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.2.4. May work with newer versions.




Accord.NET Framework 2.6.0 release notes
-----------------------------------------
05.04.2012.    

Version updates and fixes:

* General
  - Introducing experimental NuGet packages.

* Accord.Statistics
  - Adding class features to HCRF potential functions;
  - Adding resilient gradient (RProp) learning for HCRFs;
  - Adding notifications for HMM classifier learning;
  - Adding non-stochastic mode for GradientDescentHiddenLearning;
  - Adding notifications for stochastic and conjugate gradient learning;
  - Adding optimizations for HMM-based linear-chain HCRFs;
  - Implementing ICloneable in all CRF/HCRF related classes;
  - Updating positive definite checks in Cholesky decompositions;
  - Correcting HMM computations to normalize probabilities only if required;
  - Correcting DistributionBase to be marked as serializable;
  - Correcting catastrophic cancellation in HCRF evaluation;
  - Correcting output feature marginal probability calculation;
  - Correcting Phi and ChiSquare calculation in GeneralConfusionMatrix;
  - Renaming BaseSequenceClassifierLearning to BaseHiddenMarkovClassifierLearning.

* Accord.MachineLearning
  - Adding pooled variance support for Cross-Validation;
  - Adding a switch to disable parallel computations in Cross-Validation;
  - Adding support for external initialization in  Cross-Validation;
  - Adding support for compact form linear SVMs;
  - Adding kernel function cache in SVMs and SMO;
  - Adding notification events to multi-class SVM learning;
  - Adding Keerthi et al. improvements to the SMO algorithm;
  - Splitting Cross-Validation and Grid-Search classes in different files;
  - Splitting Multiquadric and InverseMultiquadric kernels;
  - Correcting Multiquadric kernel to be positive-definite;
  - Optimizing multi-class SVM evaluation based on DAGs.

* Accord.Math
  - Adding the Goldfarb-Idnani solver for constrained QP problems.

* Accord.Imaging
  - Adding further options to disable gradient blending in Blend filter;
  - Adding image rectification filter for projective transformations;
  - Adding specialized RANSAC fundamental matrix estimator.

* Sample applications
  - Adding Solver sample application for constrained optimization problems.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.2.4. May work with newer versions.




Accord.NET Framework 2.5.0 release notes
-----------------------------------------
20.01.2012.    

Version updates and fixes:

* Accord.Math
  - Splitting Special class into smaller classes;
  - Moving ComplexMatrix to ComplexExtensions namespace.

* Accord.Statistics
  - Introducing Accord.Statistics.Links namespace for link functions;
  - Adding Absolute, Cauchit, Identity, Inverse, Inverse Squared,
    Logit, Log, Log-log, Probit and Sin link functions;
  - Adding IFittableDistribution interface to eventually replace
    the functionality of IDistribution.Fit in a future version;
  - Adding Gamma, Cauchy, InverseGaussian, Hypergeometric, Geometric,
    Exponential, Binomial and Kolmogorov-Smirnov probability distributions;
  - Adding InverseDistributionFunction to most Univariate distributions;
  - Adding ComplementaryDistributionFunction for all IDistributions;
  - Adding HypothesisTest interface for specifying statistical tests;
  - Adding Kolmogorov-Smirnov tests for one and two samples;
  - Adding Student's T-Test for one and two samples;
  - Adding Quartiles function to statistical tools;
  - Adding several association measures to general contingency tables;
  - Adding Bhapkar, Bowker, Kappa, McNemar, Stuart-Maxwell and Two-matrix Kappa tests;
  - Adding symmetric triangle, Squared sinc, Hypersecant and Dirichlet kernels;
  - Adding Gamma formulations to sparse Gaussian and sparse Laplacian kernels;
  - Correcting property name from LogEmissions to Emissions in Markov models.

* Accord.MachineLearning
  - Adding small optimizations to SMO such as index sets and diagonal caching;
  - Adding support vector sharing in the computation of multi-class SVMs;
  - Adding Estimate methods for Gaussian, Laplacian and Sigmoid kernels;
  - Adding probabilistic output learning and link functions to SVMs;
  - Adding elimination-based multi-class decision for multi-class SVMs (DAGs);
  - Changing default value for constant in linear and polynomial kernels to 1;
  - Changing default behavior of SVM compute in order to return decisions.

* Accord.Neuro
  - Adding linear activation function.

* Accord.Audio
  - Adding simple mode to Wavechart control;
  - Adding navigation support to WaveEncoder;
  - Correcting Wave Encoder and Decoder classes.

* Accord.Imaging
  - Adding better starting values in Image/Matrix converters.

* Sample applications
  - Updating Handwriting (SVM) sample application to use all samples
    available and to perform automatic estimation of Gaussian kernel;
  - Adding new audio recording sample application.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.2.3. May work with newer versions.




Accord.NET Framework 2.4.0 release notes
-----------------------------------------
04.01.2012.    

Version updates and fixes:

* General
  - Updating all projects to .NET 4.0;
  - Updating all libraries to use Task Parallel Library;
  - Updating SlimDX to latest version (September 2011);
  - Adding new project for GPL source codes.

* Accord.Audio
  - Adding WaveEncoder class for saving/encoding Wave files.

* Accord.Controls
  - Updating ArrowDataView to support row labels.

Accord.Imaging
  - Correcting Blending filter and Harris corners detector convolutions.

* Accord.MachineLearning
  - Adding Nave Bayes classifier;
  - Adding Decision Trees models;
  - Adding ID3 and C4.5 learning algorithms for decision trees;
  - Adding Load and Save methods to all Support Vector Machines;
  - Correcting race condition in multi-class vector machine decision;
  - Renaming KMean's method "Nearest" to "Classify".

* Accord.Math
  - Adding Conjugate Gradient (CG) optimization algorithm;
  - Adding IsUpperTriangular, IsLowerTriangular, IsDiagonal methods to matrix class;
  - Adding Mode for integers and Entropy calculation in Math.Tools;
  - Adding Lambda Comparer for specifying ad-hoc comparison rules;
  - Adding LogSum function for summing exponential without losing accuracy;
  - Adding efficient trace of the inverse in Cholesky decomposition;
  - Adding jagged-array variants of Cholesky and Lu decompositions;
  - Adding single-precision and decimal variants of Cholesky, Qr,
    Eigenvalue, Singular value and Lu matrix decompositions;
  - Correcting matrix formatter to use argument culture;
  - Correcting regularized Gamma special functions;
  - Updating numeric decompositions to use T4 templates;
  - Renaming LbfgsOptimization to BroydenFletcherGoldfarbShanno;
  - Adding parallelization to LDLt Cholesky Decomposition.

* Accord.Neuro
  - Heavily optimizing Levenberg-Marquardt to lessen memory and CPU requirements;
  - Adding partial Jacobian calculation in Levenberg-Marquardt to conserve memory;
  - Adding Resilient Backpropagation (RProp) learning algorithm;
  - Renaming NguyenWidrowInitializer to NguyenWidrow.
  
* Accord.Statistics
  - Adding Conditional Random Fields (CRFs);
  - Adding Hidden Conditional Random Fields (HCRFs);
  - Adding Stochastic Gradient Descent (SGD) learning for HCRFs;
  - Adding Conjugate Gradient and QuasiNewton (L-BFGS) learning for HCRFs;
  - Renaming SequenceClassifiers to HiddenMarkovClassifiers;
  - Adding class priors to HiddenMarkovClassifiers;
  - Adding InverseGaussian, Nakagami, Rayleigh and Weibull distributions;
  - Adding Joint and Independent probability distributions;
  - Adding fitting options for GeneralDiscreteDistribution;
  - Adding LogProbabilityFunction method to all probability distributions;
  - Adding Load and Save methods for all Hidden Markov Models and Fields;
  - Updating Markov models to use log-domain computations by default;
  - Updating Normal fitting options to limit to diagonal covariance matrices; 
  - Correcting weighted mixture distribution fitting;
  - Correcting running normal statistics;
  - Correcting LDA classification function;
  - Correcting Kernel matrix centering in KPCA;
  - Correcting Backward algorithm for HMMs.

* Accord.Vision
  - Adding smoothness to Camshift image extraction;
  - Adding parallelism to Haar Object Detector;
  - Optimizing IntegralImage computation with unsafe operations.

* Sample applications
  - Adding Naive Bayes sample application;
  - Adding Decision Tree sample application;
  - Adding Classification sample application for
    Levenberg-Marquardt trained Neural Networks;
  - Adding XOR, TimeSeries and Approximation samples
    for ResilientBackpropagation trained Networks.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.2.3. May work with newer versions.




Accord.NET Framework 2.3.0 release notes
-----------------------------------------
01.09.2011.    

Version updates and fixes:

* General
  - Ensuring copyright information stay as complete as possible.
  
* Accord.Neuro
  - Levenberg-Marquardt now supports networks with multiple hidden layers.

* Accord.Statistics
  - Adding ANOVA and Two-way ANOVA;
  - Adding Odds Ratio, Overall Diagnostic Power and Normalized 
    Mutual Information properties for 2 x 2 confusion matrices;
  - Adding class for General Confusion Matrices (i.e. NxN confusion matrices);
  - Adding Multinomial Logistic Regression using a Lower-Bound Newton optimization;
  - Adding Predict method for arbitrary-density hidden Markov models;
  - Adding Empirical, Lognormal, T and Uniform distributions;
  - Splitting logistic regression models and related 
    fitting algorithms into independent classes;
  - Fixing Clone operation of Mixture distributions.
  - Fixing Normal distributions constructor's parameters;
  - Fixing variance calculation for multivariate mixture;
  - Replacing continuous-density hidden Markov models with generic, arbitrary-density models;
  - Correcting an infinite loop when regularization is zero in MultivariateNormalDistribution;
  - Correcting KDA to avoid argument out of range exceptions when removing unneeded components;
  - Correcting probability mass function for general discrete distribution;
  - Renaming Multivariate distributions to make a better distinction from univariate ones;
  - Renaming UniformDistribution to ContinuousUniformDistribution;
  - Updating univariate Normal to use log computations instead;
  - Adding initial class for Scatter plots.

* Accord.Math
  - Adding Kronecker product to Matrix class.

* Accord.Controls
  - Introducing ScatterplotView.

* Sample applications
  - Adding Kinect controller for interfacing with Microsoft Kinect.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.2.1. May work with newer versions.




Accord.NET Framework 2.2.0 release notes
-----------------------------------------
01.07.2011.    

Version updates and fixes:

* General
  - This release introduces some breaking interface changes. As
    result, the current minor version number has been increased;
  - Updating DimensionMismatchException to be an ArgumentException;
  - Merging Accord.Wavelets into Accord.Math.

* Accord.MachineLearning
  - Adding more argument checks in SVM learning related classes;
  - Adding a overload to initialize GMM with known distributions;
  - Adding a overload to compute K-Means errors directly from the data.

* Accord.Math  
  - Adding 1D convolution extension method in Matrix class;
  - Adding modified Bessel functions in the Special functions class;
  - Adding Finite Differences class for automatic gradient computation;
  - Adding GetAngle(IntPoint, IntPoint) and GetDirection in Math.Tools.
  - Adding new LogDeteminant properties to Cholesky and LU decompositions
    to compute the determinant's logarithm avoiding overflown values;
  - Correcting LU decomposition of matrices with more columns than rows;
  - Correcting matrix Power operation.

* Accord.Statistics
  - Adding Circular statistics tools;
  - Adding Von-Mises probability distribution;
  - Adding static methods for NormalDistribution estimation;
  - Adding Moving and Running namespaces for moving window and running statistics;
  - Adding MovingNormalStatistics class for computing moving averages and variances;
  - Adding MovingCircularStatistics class for moving circular averages and variances;
  - Adding more method overloads for float input values in Statistics.Tools;
  - Adding versions of the Forward and Backward algorithms without scaling;
  - Adding smoothing support in hidden Markov Threshold model creation;
  - Adding more argument checks in statistical analysis and models;
  - Adding base interfaces for statistical analyses;
  - Updating analyses classes to implement base interfaces;
  - Updating SequenceClassifierBase to compute the threshold model in parallel;
  - Updating Continuous Baum-Welch to automatically convert univariate distributions;
  - Updating multivariate NormalDistribution to use logarithmic computations;
  - Fixing initialization of the symbols property in Hidden Markov Model constructor;
  - Fixing univariate mixture distribution to respect component's fitting options;
  - Fixing threshold model generation in discrete sequence Markov classifiers;
  - Experimental version of linear-chain Conditional Random Fields;
  - Marking all kernel classes as sealed.
  
* Accord.Neuro
  - Correcting iterator bug in Levenberg Marquardt learning.
  
* Accord.Imaging
  - Adding Raw and Central image moments calculators;
  - Adding border following contour extraction algorithm;
  - Adding K-curvature algorithm for contour peak detection;
  - Adding convex hull defects extraction algorithm;
  - Updating PointsMarker and RectanglesMarker to accept
    any IEnumerable collection rather than just arrays.

* Accord.Vision
  - Adding HSL Range Blob Tracker for color-based tracking;
  - Adding Block Matching Tracker for template-based tracking;
  - Adding steady detection in Camshift and HaarObjectDetector;
  - Adding new classes for default Face and Nose Haar cascades;
  - Adding support for hard-coded code generation in Haar cascades;
  - Adding Axis orientation support for TrackingObject;
  - Updating TrackingObject to use integer coordinates;
  - Updating all Haar related classes to implement the ICloneable interface;
  - Splitting HaarCascade class from the HaarCascadeClassifier;
  - Fixing multi-threaded issues when using the default face detector;
  - Fixing red channel extraction in HaarObjectDetector;
  - Fixing minimum window size in HaarObjectDetector.
  
* Accord.Controls
  - Adding Accord.Controls.Vision namespace for vision related controls;
  - Adding AngleBox and PointBox controls to show angle and position information;
  - Adding HeadController component for interface control based on head events;
  - Adding FaceController component for interface control based on face events;

* Sample applications
  - Adding Controller sample application for head and face based application control;
  - Adding Glove Tracking sample application for colored glove tracking and finger detection;
  - Updating Mouse Gestures sample application to perform continuous recognition using
    hidden Markov threshold models and use discrete models instead.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.1.5. May work with newer versions.




Accord.NET Framework 2.1.6 release notes
-----------------------------------------
01.04.2011.    

Version updates and fixes:

* General
  - Introducing new exception classes for math operations.

* Accord.Audio
  - Adding new AudioDeviceCollection class to list available 
    audio devices information based on category (output or capture);
  - Updating AudioCaptureDevice and AudioOutputDevice to actually 
    support any available audio device (Origo Issue #21).
 
* Accord.MachineLearning
  - Adding methods to (re-)initialize Gaussian Mixture Models and K-Means;
  - Adding support for setting the centroids in K-Means (Origo Issue #20);

* Accord.Math
  - Adding L-BFGS method for non-linear optimization;
  - Adding optimized matrix inversions for 3x3 and 2x2 matrices; 
  - Matrix decompositions now implement the ICloneable interface;
  - Updating Multivariate/NormalDistribution to always use the
    Cholesky decomposition and throw exceptions when appropriate;
  - Updating Cholesky decomposition to cache the result of the
    determinant property to avoid unnecessary computations.

* Accord.Statistics
  - New architecture for statistical distribution fitting;
  - Adding a regularization parameter in multivariate Normal distribution
    fitting to avoid non-positive definite or singular covariance matrices;
  - Adding configurable fitting options in the ContinuousBaumWelchLearning;
  - Adding confidence support in sequence classifiers using Threshold models;
  - Adding support for inclusive upper bounds in the Histogram class;
  - Adding more selection rules for the number and widths of Histogram bins;
  - Fixing Histogram computing when data is constant (Origo Issue #23);
  - Renaming weighted sample methods from Measures.
  
* Accord.Imaging
  - Adding Speeded-Up Robust Features (SURF) detector and descriptor;
  - Adding Features from Accelerated Segment Test (FAST) corners detector;

* Sample applications
  - Mouse gestures sample application now supports non-gesture models
    and a regularization parameter for Normal distribution estimation.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.1.5. May work with newer versions.




Accord.NET Framework 2.1.5 release notes
-----------------------------------------
12.02.2011.    

Version updates and fixes:

* General
  - The various works on this release have introduced some breaking
    interface changes, mainly in the Audio and Statistics namespaces.

* Accord.Audio
  - Major architectural changes of the Audio namespace;
  - Code cleanup of untested audio processing filters;
  - Signal processing filters now operate in more similar
    ways to AForge.NET's image processing filters;
  - Signals can now have multiple sample formats.
 
* Accord.MachineLearning
  - Updating GridSearch to use parallel computing;
  - Adding a new option to automatically compute the complexity parameter
    C using a heuristic in Sequential Minimal Optimization SVM learning.

* Accord.Math
  - Reorganizing Special functions, adding (ln) Beta functions;
  - Splitting Matrix.cs into several partial classes for organization;
  - Adding MultiplyByTranspose and TransposeAndMultiply methods for matrices;
  - Adding a new algorithm for LDLt Cholesky matrix decomposition;
  - Adding a new property in SVD to store original value sorting order;
  - Adding options for Eigenvalue and Singular value decompositions to
    be computed in place, conserving memory.

* Accord.Neuro
  - Updating LevenbergMarquardtLearning to introduce a constant factor
    while computing the Hessian/Bayesian Regularization weight update.

* Accord.Statistics
  - Major refactoring of the Statistics.Models.Markov namespace;
  - Adding a new Predict method in HMMs to predict next observations in a sequence;
  - Adding Topology classes to specify a Hidden Markov Model architecture at model
    construction, such as Ergodic (fully connected) or Forward-only (left-to-right);
  - Adding a base class and interfaces for Continuous and Discrete hidden Markov models;
  - Adding a base class and interfaces for hidden Markov model learning algorithms;
  - Adding a class for the Univariate F-Distribution;
  - Adding a class for Independent Component Analysis (ICA);
  - Adding a Whitening Transform method in Statistics.Tools;
  - Adding Sparse versions of Gaussian, Polynomial, Laplacian, Sigmoid
    and Cauchy kernels to operate on sparse vectors directly;
  - Adding options for Principal Component Analysis and Kernel Principal
    Component Analysis to be computed in place, conserving memory;
  - Updating Projection filter to use a Collection instead of List<T> (CA1002);
  - Updating Histogram and HistogramView to work with empty sample vectors;
  - Updating Principal Component Analysis and Partial Least Squares Analysis
    to raise exception if std deviation is zero, instead of failing silently;
  - Updating Evaluate method in Hidden Markov models to compute forward
    probabilities instead of the probability of the most likely (Viterbi) path;
  - Fixing an issue with Mixture model fitting when one of the weights is zero.
  

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.1.5. May work with newer versions.




Accord.NET Framework 2.1.4 release notes
-----------------------------------------
01.12.2010.    

Version updates and fixes:
 
* Accord.Core
  - Adding a new assembly, Accord.Core, to accommodate common
    framework classes such as specialized exceptions;

* Accord.Math
  - Adding some Tools.Scale overloads to operate on matrices and arrays;
  - Adding an option to compute LU decomposition and transpose in place;
  - Cleaning Matrix to have only generic implementations for Max and Min;
  - Fixing a bug in Nonnegative Matrix Factorization in which certain
    input characteristics would raise an IndexOutOfRange exception.

* Accord.MachineLearning
  - Adding a SparseSampleReader class to read from files containing
    sparse data samples represented in libsvm's sparse vector format;
  - Changing MulticlassSupportVectorMachine to implement the interface
    ISupportVectorMachine.Compute explicitly, so the method can also
    return a more intuitive integer class label rather than a double.
  
* Accord.Neuro
  - Adding several optimizations to Levenberg-Marquardt learning.

* Accord.Statistics
  - Adding a Custom (lambda) kernel for ad-hoc kernels;
  - Adding a Precomputed kernel for manually specifying a kernel Matrix;
  - Adding a Sparse Linear Kernel to operate on sparse vectors directly;
  - Adding support for different weights in the Additive combination kernel;
  - Adding support for setting the Gamma parameter in the Gaussian kernel;
  - Moving nested classes in StepwiseLogisticRegression to separate locations;
  - Correcting error calculation in Multiple Linear Regression.

* Accord.Imaging
  - Adding a ToBitmap overload to convert from byte matrices;
  - Adding a new overload to the Scale method to accept floats;
  - Adding a new overload to ImageBox to support background color;
  - Correcting Harris Corners Detector (with support for Nobel's corner measure);
  - Correcting Blend filter to ignore fully transparent pixels;
  - Modifying Blend filter to use float instead of doubles.

* Accord.Vision
  - Adding different scaling modes to Haar Cascade Object Detector;
  - Adding initial support for Tilted features in HaarCascadeClassifier and
    Tilted Integral Image calculation in IntegralImage2 class;
  - Rewrite of the Haar Cascade Classifier to work with all compatible XML
    definitions. Serialization is now performed using standard .NET serializers.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.1.4. May work with newer versions.




Accord.NET Framework 2.1.3 release notes
-----------------------------------------
01.11.2010.    

Version updates and fixes:
 
* General
  - Heavy work on documentation;
  - Adding parameter verification in several methods;

* Accord.Math
  - Adding Non-Negative Matrix Factorization;
  - Adding a new Split method to split vectors in Matrix class;
  - Adding a new IsPositiveDefinite method in Matrix class;
  - Adding a new method for equation solving in LU and QR
    decompositions to help in multiplication by inverse;
  - Adding a new overload to the Solve method in Cholesky
    decomposition to solve for vectors;
  - Updating Matrix.Divide to use the LU and QR decompositions;
  - Correcting a bug in the Submatrix method for vectors;
  - Reworking some of the matrix formatting architecture.

* Accord.MachineLearning
  - Adding Gaussian Mixture Models;
  - Adding configurable distance metric and adjustable/relative
    convergence threshold to K-Means clustering;
  - Updating the interface of Crossvalidation and Multiclass Support
    Vector Machines to use delegates instead of raw lambda functions.

* Accord.Statistics
  - Major reorganization of the Statistics.Markov namespace;
  - Major reorganization of the Statistics.Distributions namespace;
  - Adding Continuous Hidden Markov Models and Sequence Classifier;
  - Adding support for univariate/multivariate Mixture distributions;
  - Adding Poisson, Multinomial and Bernoulli probability distributions; 
  - Adding overloads to Statistics.Tools for weighted measures;
  - Correcting an infinite loop issue in hidden Markov models learning;
  - Fixing Wavelet kernel - translation parameter was being ignored.

* Accord.Imaging
  - Adding ToBitmap/ToDoubleArray extension methods for color images;
  - Correcting a bug in distance function in RansacHomograpyEstimation.

* Accord.Vision
  - Adding HaarCascadeClassifier as a parameter to HaarObjectDetector;
  - Changing DefaultFaceDetector to be a property rather than a method.

* Accord.Controls
  - Adding a RangeTypeConverter to show AForge.IntRange and DoubleRanges
    properly in controls such as the PropertyGrid.

* Samples 
  - Added Continuous density Hidden Markov Models sample application;
  - Added Gaussian Mixture Models sample application;
  - Added K-Means color clustering sample application.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.1.4. May work with newer versions.




Accord.NET Framework 2.1.2 release notes
-----------------------------------------
01.10.2010.    

Version updates and fixes:
 
* Accord.Math
  - Adding formatting and parsing extension methods for matrices;
  - Fixing an infinite-loop condition in SVD, also updating the
    inverse and solve methods to use optimized matrix multiplication;

* Accord.MachineLearning
  - Adding initial version of the K-Means clustering algorithm;

* Accord.Statistics
  - Adding a missing overload of the Variance method;
  - Adding a minimum variance threshold for components in KPCA;
  - Fixing an issue with the Result property of PCA, KPCA, LDA and KDA;

* Accord.Imaging
  - Adding ToBitmap extension methods;
  - Performance improvements in the Harris Corners Detector;
  - Fixing interface compatibility with the latest version of AForge.
    Minimum required version of AForge.NET is now version 2.1.4.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.1.4. May work with newer versions.




Accord.NET Framework 2.1.1 release notes
-----------------------------------------
01.09.2010.    

Version updates and fixes:

* General
  - Most model classes are now serializable;
  - Adding a new icon for the framework;
  - Adding a great number of unit tests;
  - Several improvements in documentation;

* Accord.Vision
  - Fixing some issues with Camshift in HSL mode;
  - Adding a Camshift mixed mode to work on a HSL space
    with some lightness/saturation information;

* Accord.Math
  - Added optimizations to some matrix operations;
  - Added autocorrection for JAMA's SVD assumptions;
  - Added more tests for the matrix decompositions;

* Accord.MachineLearning
  - Corrected the casing of the name of GridSearch related classes;

* Accord.Statistics
  - Several corrections to basic statistics tools;
  - Added Wavelet and Dynamic Time Warp kernels;

* Accord.Imaging
  - Added support for 32bpp images in the blending
    filter, also fixed a issue with final image size;
  - Some fixes for the Correlation Matching class;

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.1.3. May work with other versions.




Accord.NET Framework 2.1.0 release notes
-----------------------------------------
01.08.2010.    

Version updates and fixes:

* General
  - Updated AForge version to v2.1.3
  - Updating some projects to target .NET 3.5
  - Added the Accord.Vision namespace

* Accord.Statistics
  - Correction of the ANOVA kernel;
  - Added minor optimizations to some radial basis function kernels;
  - Added option to force the inclusion of the origin point in ROC curves;

* Accord.Imaging
  - Fixed blending filter issue when second image was placed at left of the first image.
  - WaveletTransform now works properly on 16bpp Grayscale images for added precision;

* Accord.Vision
  - Added support for Camshift-based object tracking;
  - Added support for the Viola-Jones method for face detection;

* Accord.Math
  - Added more comments on the Singular Value Decomposition issues;
  - Added more extension methods to Accord.Math.Matrix, such as CartesianProduct;
  - Updated Factorial, Binomial and other special functions;

* Accord.MachineLearning
  - Added support for Multiclass (Kernel) Support Vector Machines;
  - Added a Grid Search class for parameter tuning;
  - Added a Crossvalidation class for model measurement;
  - Sequential Minimal Optimization now works with common (linear) machines;

* Samples 
  - Adjusting Panorama sample application;
  - Removed incomplete SignalSplitter sample;
  - Added Multiclass Support Vector Machine-based Handwriting Recognition sample application;
  - Added sample applications for face tracking and detection
  - Added a sample application to demonstrate some matrix operations.

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.1.3. May work with other versions.




Accord.NET Framework 2.0.2 release notes
-----------------------------------------
01.07.2010.            

Version updates and fixes:

* General
  - Some refactoring to comply with more Code Analysis rules

* Accord.Statistics
  - Added common base class for DataTable Filters
  - Added computation of Variable Importance in Projection (VIP) for PLS
  - Changed Kernel Discriminant Analysis to use a Generalized Eigenvalue
    Decomposition for increased efficiency.

* Accord.Imaging
  - Added Wavelet transform filter
  - Improvements to the Blending filter
  - Adding sample pictures for the Panorama sample application
  - Added MatrixH to represent general transformation matrices more efficiently
  - Simplification of RansacHomographyEstimator using PointF instead of PointH

* Accord.MachineLearning
  - Adding initial Crossvalidation class for general models
  - Simplifying RANSAC to be independent of the model data type

* Accord.Wavelets
  - Updating the Wavelet libraries and adding a sample application

* Accord.Math
  - Added special functions Log1p and Expm1 to Accord.Math.Special
  - Added support for computing Generalized Eigenvalue Decompositions

* AForge Compatibility
  - Compiled against AForge.NET Framework 2.1.2. May work with other versions.




Accord.NET Framework 2.0.1 release notes
-----------------------------------------
01.06.2010.            

Version updates and fixes:

* General
  - Added support for Sandcastle generated documentation
  - Added unit testing for Imaging, MachineLearning, Math, Neuro, Statistics and Wavelets namespaces
  - Updated assemblies information

* Accord.Statistics
  - Fixed an issue with Kernel Principal Component Analysis reversion

* Accord.Wavelets
  - Added Haar and CDF97 1D and 2D Wavelet Transforms




Accord.NET Framework 2.0.0 release notes
-----------------------------------------
20.05.2010.

Initial release.

* Accord.Math
  - MATLAB-like matrix manipulation of multidimensional-arrays
  - Hilbert transform
  - Numeric decompositions (SVD, Eigen, Cholesky, LU, QR)
  - Special scientific functions;

* Accord.Statistics
  - Simple Descriptive Analysis
  - Linear Discriminant Analysis (LDA)
  - Kernel Discriminant Analysis (KDA)
  - Principal Component Analysis (PCA)
  - Kernel Principal Component Analysis (KPCA)
  - Partial Least Squares Analysis (PLS)
  - Logistic Regression Fitting and Analysis
  - Multivariate, Multiple and Simple Linear Regressions
  - Receiver Operating Curves (ROC)
  - Chi-Square and Wald Hypothesis Tests
  - Wide variety of Kernels for machine learning
  - Hidden Markov Models (HMM) and Sequence Classifiers
  - Filters for preprocessing of DataTables such as
    discretization, normalization, categorization and equalization
  
* Accord.Neuro
  - Levenberg-Marquardt (with Bayesian Regularization) Learning Algorithm
  - Nguyen-Widrow Weight Initialization for Activation Networks
  
* Accord.MachineLearning
  - Support Vector Machines (SVM)
  - Kernel Support Vector Machines (kSVM)
  - Sequential Minimal Optimization Learning (SMO)
  - Sequential Minimal Optimization for Regression (SMOr)
  - Multipurpose RANSAC for robust model estimation

* Accord.Imaging
  - Harris Corner Detector for AForge.NET
  - Robust Homography estimation using RANSAC
  
* Accord.Audio
  - Experimental audio library for audio processing
  
* Accord.Audition
  - Experimental audio library for computer audition

  
* Samples
  - Sample applications for almost all features listed here.