About: The apcluster package implements Frey's and Dueck's Affinity Propagation clustering in R. The package further provides leveraged affinity propagation, exemplar-based agglomerative clustering, and various tools for visual analysis of clustering results.

• plotting of clustering results extended to data sets with more than two dimensions (resulting in the clustering result being superimposed in a scatterplot matrix); the variant that plot() can be used to draw a heatmap has been removed. From now on, heatmap() must always be used.
• improved NA handling
• correction of input check in apcluster() and apclusterL() (previously, both functions issued a warning whenever argument p had length > 1)
• corresponding updates and further improvements of help pages and vignette

About: The Cognitive Foundry is a modular Java software library of machine learning components and algorithms designed for research and applications.

• General:
  • Made code able to compile under both Java 1.6 and 1.7. This required removing some potentially unsafe methods that used varargs with generics.
  • Upgraded XStream dependency to 1.4.4.
  • Improved support for regression algorithms in learning.
  • Added general-purpose adapters to make it easier to compose learning algorithms and adapt their input or output.
• Common Core:
  • Added isSparse, toArray, dotDivide, and dotDivideEquals methods for Vector and Matrix.
  • Added scaledPlus, scaledPlusEquals, scaledMinus, and scaledMinusEquals to Ring (and thus Vector and Matrix) for potentially faster such operations.
  • Fixed issue where matrix and dense vector equals was not checking for equal dimensionality.
  • Added transform, transformEquals, tranformNonZeros, and transformNonZerosEquals to Vector.
  • Made LogNumber into a signed version of a log number and moved the prior unsigned implementation into UnsignedLogNumber.
  • Added EuclideanRing interface that provides methods for times, timesEquals, divide, and divideEquals. Also added Field interface that provides methods for inverse and inverseEquals. These interfaces are now implemented by the appropriate number classes such as ComplexNumber, MutableInteger, MutableLong, MutableDouble, LogNumber, and UnsignedLogNumber.
  • Added interface for Indexer and DefaultIndexer implementation for creating a zero-based indexing of values.
  • Added interfaces for MatrixFactoryContainer and DivergenceFunctionContainer.
  • Added ReversibleEvaluator, which various identity functions implement as well as a new utility class ForwardReverseEvaluatorPair to create a reversible evaluator from a pair of other evaluators.
  • Added method to create an ArrayList from a pair of values in CollectionUtil.
  • ArgumentChecker now properly throws assertion errors for NaN values. Also added checks for long types.
  • Fixed handling of Infinity in subtraction for LogMath.
  • Fixed issue with angle method that would cause a NaN if cosine had a rounding error.
  • Added new createMatrix methods to MatrixFactory that initializes the Matrix with the given value.
  • Added copy, reverse, and isEmpty methods for several array types to ArrayUtil.
  • Added utility methods for creating a HashMap, LinkedHashMap, HashSet, or LinkedHashSet with an expected size to CollectionUtil.
  • Added getFirst and getLast methods for List types to CollectionUtil.
  • Removed some calls to System.out and Exception.printStackTrace.
• Common Data:
  • Added create method for IdentityDataConverter.
  • ReversibleDataConverter now is an extension of ReversibleEvaluator.
• Learning Core:
  • Added general learner transformation capability to make it easier to adapt and compose algorithms. InputOutputTransformedBatchLearner provides this capability for supervised learning algorithms by composing together a triplet. CompositeBatchLearnerPair does it for a pair of algorithms.
  • Added a constant and identity learners.
  • Added Chebyshev, Identity, and Minkowski distance metrics.
  • Added methods to DatasetUtil to get the output values for a dataset and to compute the sum of weights.
  • Made generics more permissive for supervised cost functions.
  • Added ClusterDistanceEvaluator for taking a clustering that encodes the distance from an input value to all clusters and returns the result as a vector.
  • Fixed potential round-off issue in decision tree splitter.
  • Added random subspace technique, implemented in RandomSubspace.
  • Separated functionality from LinearFunction into IdentityScalarFunction. LinearFunction by default is the same, but has parameters that can change the slope and offset of the function.
  • Default squashing function for GeneralizedLinearModel and DifferentiableGeneralizedLinearModel is now a linear function instead of an atan function.
  • Added a weighted estimator for the Poisson distribution.
  • Added Regressor interface for evaluators that are the output of (single-output) regression learning algorithms. Existing such evaluators have been updated to implement this interface.
  • Added support for regression ensembles including additive and averaging ensembles with and without weights. Added a learner for regression bagging in BaggingRegressionLearner.
  • Added a simple univariate regression class in UnivariateLinearRegression.
  • MultivariateDecorrelator now is a VectorInputEvaluator and VectorOutputEvaluator.
  • Added bias term to PrimalEstimatedSubGradient.
• Text Core:
  • Fixed issue with the start position for tokens from LetterNumberTokenizer being off by one except for the first one.

About: This toolbox implements models for Bayesian mixed-effects inference on classification performance in hierarchical classification analyses.

In addition to the existing MATLAB implementation, the toolbox now also contains an R package of the variational Bayesian algorithm for mixed-effects inference.

About: The aim is to embed a given data relationship matrix into a low-dimensional Euclidean space such that the point distances / distance ranks correlate best with the original input relationships. Input relationships may be given as (asymmetric) distances, dissimilarities, or (negative) scores. Input-output relations are modelled as row-conditioned. (Weighted) Pearson and soft Spearman rank correlation, and unweighted soft Kendall correlation are supported correlation measures for input/output object neighborhood relationships.

• Initial release (Ver 1.0): Weighted Pearson and correlation and soft Spearman rank correlation, Tue Dec 4 16:14:51 CET 2012

• Ver 1.1 Added soft Kendall correlation, Fri Mar 8 08:41:09 CET 2013

About: LIBOL is an open-source library that consists of a family of state-of-the-art online learning algorithms for machine learning and data mining research.

Initial Announcement on mloss.org.

About: FABIA is a biclustering algorithm that clusters rows and columns of a matrix simultaneously. Consequently, members of a row cluster are similar to each other on a subset of columns and, analogously, members of a column cluster are similar to each other on a subset of rows. Biclusters are found by factor analysis where both the factors and the loading matrix are sparse. FABIA is a multiplicative model that extracts linear dependencies between samples and feature patterns. Applications include detection of transcriptional modules in gene expression data and identification of haplotypes/>identity by descent< consisting of rare variants obtained by next generation sequencing.

CHANGES IN VERSION 2.4.0

o spfabia bugfixes

CHANGES IN VERSION 2.3.1

NEW FEATURES

o Getters and setters for class Factorization

2.0.0:

• spfabia: fabia for a sparse data matrix (in sparse matrix format) and sparse vector/matrix computations in the code to speed up computations. spfabia applications: (a) detecting >identity by descent< in next generation sequencing data with rare variants, (b) detecting >shared haplotypes< in disease studies based on next generation sequencing data with rare variants;
• fabia for non-negative factorization (parameter: non_negative);
• changed to C and removed dependencies to Rcpp;
• improved update for lambda (alpha should be smaller, e.g. 0.03);
• introduced maximal number of row elements (lL);
• introduced cycle bL when upper bounds nL or lL are effective;
• reduced computational complexity;
• bug fixes: (a) update formula for lambda: tighter approximation, (b) corrected inverse of the conditional covariance matrix of z;

1.4.0:

• New option nL: maximal number of biclusters per row element;
• Sort biclusters according to information content;
• Improved and extended preprocessing;
• Update to R2.13

About: ELKI is a framework for implementing data-mining algorithms with support for index structures, that includes a wide variety of clustering and outlier detection methods.

This is mostly a bug fix release. A lot of small issues have been fixed that improve performance, make error reporting a lot better, ease the use of sparse vectors and external precomputed distances, for example.

This will be the last ELKI release to support Java 6. The next ELKI release will require Java 7.

Algorithms

• Some new LOF variants (LDF, SimpleLOF, SimpleKernelDensityLOF)
• Correlation Outlier Probabilities (ICDM 2012)
• A naive mean-shift clustering
• Single-link clustering (SLINK algorithm) should be significantly faster due to optimized data structures
• "Benchmarking" algorithms for measuring the performance of index structures

Index layer

• Bulk loading R-Trees should be faster - in particular Sort Tile Recursive can work very well.
• M-Trees have been refactored and optimized for double distances

Database layer

• Bundle format (work in progress): low-level binary format for fast data exchange
• DBID and DataStore layer received some additional classes for further performance improvements
• KNN heap structures were revisited. The code is less clean now, but performs better in benchmarks.

Visualizations

• General clean up and API simplifications
• Some additional modules and improvements

Various

• There is a new parameter class, RandomParameter
• Some new distributions were added, also to the data set generator.

Tutorials

• The website has new tutorials, including one on a k-means variation that produces equal sized clusters.

About: Python Framework for Vector Space Modelling that can handle unlimited datasets (streamed input, online algorithms work incrementally in constant memory).

• added the "hashing trick" (by Homer Strong)
• support for adding target classes in SVMlight format (by Corrado Monti)
• fixed problems with global lemmatizer object when running in parallel on Windows
• parallelization of Wikipedia processing + added script version that lemmatizes the input documents
• added class method to initialize Dictionary from an existing corpus (by Marko Burjek)

About: Stochastic neighbor embedding aims at the reconstruction of given distance, dissimilarity, or score neighborhood relations in a low-dimensional Euclidean space. This can be regarded as general approach to multi-dimensional scaling, but the reconstruction is based on the definition of input (and output) neighborhood probability alone. Probability of score exceedance is used for neighborhood probability estimation, which is connected to soft-rank optimization. The present implementation makes use of quasi 2nd order gradient-based (l-)BFGS optimization.

• scoretoprob.m replaced by d2p.m

• protein score data set added

• trank.m computes (mid/max -tied) ranks along columns of matrix

• local P- neighborhood probability estimation added

• experimental soft_rank_SNE added for minimizing KL between probabilities of exceedance in source and embedding space

• symmetry option removed, because this was strange in previous version

About: PLEASD: A Matlab Toolbox for Structured Learning

Initial Announcement on mloss.org.

About: A general purpose library to process and predict sequences of elements using echo state networks.

Initial Announcement on mloss.org.

About: MLwizard recommends and optimizes classification algorithms based on meta-learning and is a software wizard fully integrated into RapidMiner but can be used as library as well.

Faster parameter optimization using genetic algorithm with predefined start population.

About: Matlab code for learning probabilistic SVM in the presence of uncertain labels.

Added missing dataset function (thanks to Hao Wu)

About: This package contains a python and a matlab implementation of the most widely used algorithms for multi-armed bandit problems. The purpose of this package is to provide simple environments for comparison and numerical evaluation of policies.

Initial Announcement on mloss.org.

About: The package provides a Lagrangian approach to the posterior regularization of given linear mappings. This is important in two cases, (a) when systems are under-determined and (b) when the external model for calculating the mapping is invariant to properties such as scaling. The software may be applied in cases when the external model does not provide its own regularization strategy. In addition, the package allows to rank attributes according to their distortion potential to a given linear mapping.

Version 1.1 (May 23, 2012) memory and time optimizations distderivrel.m now supports assessing the relevance of attribute pairs

Version 1.0 (Nov 9, 2011) * Initial Announcement on mloss.org.