Scribd
Upload
16 views5 pages

Apache Commons Math 4.0-beta1 Release Notes

The Apache Commons Math team released version 4.0-beta1 of the commons-math library. This is a major release that splits functionality into new components and modularizes code. It requires changes to applications that depend on commons-math. The minimum Java version is now 8 and users are encouraged to upgrade to this version and the new components.

Uploaded by

Erisa Matoshi
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as TXT, PDF, TXT or read online on Scribd
16 views5 pages

Apache Commons Math 4.0-beta1 Release Notes

The Apache Commons Math team released version 4.0-beta1 of the commons-math library. This is a major release that splits functionality into new components and modularizes code. It requires changes to applications that depend on commons-math. The minimum Java version is now 8 and users are encouraged to upgrade to this version and the new components.

Uploaded by

Erisa Matoshi
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as TXT, PDF, TXT or read online on Scribd

Apache Commons Math 4.

0-beta1 RELEASE NOTES

The Apache Commons Math team is pleased to announce the release of commons-math-
parent-4.0-beta1

The Apache Commons Math project is a library of lightweight mathematics


and statistics components addressing common practical problems.

This is a major release that requires source changes in applications.


Lower-level functionality has been split off to new components, often
with bug-fixes, performance enhancements and improved API:
Commons RNG,
Commons Numbers,
Commons Geometry,
Commons Statistics.
Functionality still within Commons Math has been modularized and
partly refactored.
This is a work-in-progress (help welcome).
Codes for which the refactoring has not been undertaken (or was
deemed incomplete) have been moved to 'legacy' modules.
The minimum version of the Java platform required to compile and use
Commons Math is Java 8.
Users are encouraged to upgrade to this version, and to the new
components listed above (they are now Commons Math's dependencies).
For code now located in the 'commons-math-legacy' module, upgrading
will be a matter of changing the top-level package name:
From '[Link].math3' to '[Link]'.
Code in the 'commons-math-legacy-core' module comprise classes that
moved from package '[Link].math3' and package
'[Link]' (e.g. 'MathArrays') to package
'[Link]'.
Notes:
Class 'FastMath' was renamed 'AccurateMath' (now located in package
'[Link]').
Caveat:
Although hundreds of issues have been fixed (see 'git log' and the
reports in the bug-tracking system of all the listed components),
many remain due to the lack of human resources necessary in order
to support the whole codebase (it was one of the main reasons for
creating more focused components).

Changes in this version include:

New features:
o MATH-1648: "BicubicInterpolator": Provide partial derivatives of interpolating
functions. Thanks to Alessandro Moscatelli.
o MATH-1371: Add "ElkanKMeansPlusPlusClusterer" clustering algorithm. Thanks to
Artem Barger.
o MATH-1622: Update class "SimplexOptimizer" (in package
"[Link]")
to support simulated annealing.
o MATH-1551: "Percentile": Partial support for weighted data. Thanks to Allen Yu.
o MATH-1529: "AkimaSplineInterpolator": Option to use alternative weights.
o MATH-1509: Add "MiniBatchKMeansClusterer" clustering algorithm. Thanks to Chen
Tao.
o MATH-1519: Add "Calinski-Harabsz" clustering evaluator. Thanks to Chen Tao.
o MATH-1516: Interface for ranking a list of clusters.
o MATH-1477: "MillerUpdatingRegression": Fixed "ArrayIndexOutOfBounds" exception.
Thanks to Chee Sing Lee.
o MATH-1459: Create a way to automatically calculate a Jacobian matrix using a
differentiator. Thanks to Adrian Porter.
o MATH-1426: "DescriptiveStatistics": allow "Double[]" as initializer. Thanks to
Karl Richter.
o MATH-1383: Dependency toward the "Commons Rng" component.
o MATH-1015: Gauss-Laguerre quadrature. Thanks to Thomas Neidhart.
o MATH-1350: Improved code coverage (unit tests). Thanks to Rob Tompkins.
o MATH-1336: New 64-bits RNG implementations. On 64-bits systems, they are ~35%
faster than
"[Link]" for generating "long" or "double" values. They also
guarantee
much better randomness than "Random" (cf. MATH-1327).
o MATH-1335: Refactoring of uniform random number generator functionality: new API
implemented in package "[Link]".
o MATH-1337: Adaptor class from new API ("[Link]") to old API
("[Link]").
o MATH-1327: Report on performance and correctness of the new RNG implementations.
o MATH-1314: Javadoc of the new RNG API ("[Link]") mentions the "seed
problem".
o MATH-1158: New interface for sampling from distribution (package
"[Link]").
o MATH-1351: New sampler API for "MultivariateRealDistribution".
o MATH-1348: Adaptor for using one of the new RNG implementations where an
instance of
"[Link]" is required.
o MATH-1295: "FactorialLog": Cache-based computation of the "log factorial"
function (implemented
as an inner class of "CombinatoricsUtils" in package "[Link]").
Thanks to Aleksei Dievskii.
o MATH-1271: Added divideUnsigned and remainderUnsigned to ArithmeticUtils. Thanks
to Qualtagh.
o Added a fast implementation of IEEEremainder in FastMath.
o MATH-945: Added method "T log10()" to interface "RealFieldElement".
o MATH-1039: Added logDensity(double) to RealDistribution and logProbability(int)
to IntegerDistribution interface. The implementations have already been
updated in release 3.3. Thanks to Aleksei Dievskii.

Fixed Bugs:
o MATH-1647: "HaltonSequenceGenerator": Raise exception when precondition is not
met. Thanks to Maksym Bohachov.
o MATH-1462: "EmpiricalDistribution": Use constant kernel for bins that contain up
to 3 values.
o MATH-1379: Fix "LoessInterpolator" (in package
"[Link]"). Thanks to Richard Wilkinson.
o MATH-1617: Class "BigReal": Fix equality check. Thanks to Ng Tsz Sum.
o MATH-1613: Decrease maximum number of iterations in "SimpsonIntegrator" and
"TrapezoidIntegrator" (due to the available range of the internal
counter of function evaluations).
o MATH-1431: "EmpiricalDistribution" handles empty bin. Thanks to Artem Onuchin.
o MATH-1565: Add context to "OutOfRangeException". Thanks to Randy Strauss.
o MATH-1558: "MidPointIntegrator": Fix iterative procedure in order to actually
benefit
from evaluations performed at earlier stages. Thanks to Sam Ritchie.
o MATH-1555: "Atan2": Documentation issue. Thanks to Laurent Galluccio.
o MATH-1549: "SimplexTableau": Internally "scale down" the problem definition when
the
constraints are defined with large numbers, in order to avoid spurious
exceptions caused by floating-point accuracy errors. Thanks to Mohammad
Rezaei.
o MATH-1548: Avoid inefficiencies in computing the standard quality measures of a
SOFM.
o MATH-1537: Clean-up (typos and unused "import" statements). Thanks to Jin Xu.
o MATH-1531: "EmpiricalDistribution": Workaround to avoid spurious exception.
o MATH-1511: Issue in site generation and developers documentation.
o MATH-1518: Remove code duplication by moving method to class "Cluster".
o MATH-1388: Avoid overflow.
o MATH-1401: "ClopperPearsonInterval": Missing cases. Thanks to Michele De
Stefano.
o MATH-1463: "[Link]": Throw "NoSuchElementException" from
"next" method.
o MATH-1458: "SimpsonIntegrator": Fixed iteration. Thanks to Alex Herbert.
o MATH-1419: "PolynomialSplineFunction": incorrect usage of exception. Thanks to
Hang Park.
o MATH-1413: add generics to the Frequency class
o MATH-1417: "RRQRDecomposition": bug in method "performHouseholderReflection".
Thanks to Jean-Francois Lecomte.
o MATH-1381: Fix "BinomialTest": P-value larger than 1. Thanks to Kexin Xie.
o MATH-1382: "MultivariateNormalDistribution": performance improvement (by
removing unnecessary copying). Thanks to Stefan Bunk.
o MATH-1405: "KolmogorovSmirnovTest": bug in "fixTies". Thanks to Daniil Finkel.
o MATH-1376: "SimplexOptimizer": Wrong value of iteration number was passed to the
convergence checker. Thanks to Thomas Weise.
o MATH-1319: Major refactoring of package "[Link]".
o MATH-1366: Removed obsolete RNG API (replaced by classes under "[Link]").
o MATH-1340: Avoid exception swallowing (in class "TestUtils").
o MATH-1339: Avoid redundant object creation.
o MATH-1356: "HypergeometricDistribution": Handle special cases that produced NaN.
o MATH-1343: "BetaDistribution": Removed unnecessary recomputation.
o MATH-1124: "JDKRandomGenerator": Method "nextInt(int)" now throws a
"NotStrictlyPositiveException".
The class now delegates to (rather inherits from) "[Link]".
o MATH-1300: "AbstractRandomGenerator" and "BitsStreamGenerator": Remove a
redundant call
to the random data provider. Thanks to Rostislav Krasny.
o MATH-1231: Lifted unnecessary restriction on constructor's argument of
"MicrosphereInterpolator" (package "[Link]").
o MATH-1230: The "SimplexSolver" will now throw a "DimensionMismatchException"
when calling "optimize(...)" with linear constraints whose dimension
does not match the dimension of the objective function.
o MATH-1222: Use [Link] rather than x != x in FastMath. Thanks to Benedikt
Ritter.
o MATH-1143: Added helper methods to FunctionUtils for univariate and multivariate
differentiable functions conversion.
o MATH-964: Removed unused package private class PollardRho in package primes.
o MATH-1221: Improve performance of "ZipfDistribution" by caching the nth
generalized harmonic.
o MATH-1205: Methods "evaluate(...)" of class "Variance" changed the internal
state
although it was stated differently in the javadoc.
o MATH-1191: Fixed ignored method parameters in QRDecomposition protected methods.
o MATH-1212: Changed javadoc as the RandomDataGenerator class does not implement
an interface anymore (the previous interface has been deprecated in
3.X and removed in 4.0).

Changes:
o MATH-1653: Remove class "CombinatoricsUtils" (in package "[Link]").
Functionality is in "Commons Numbers" combinatorics package.
o MATH-1640: Do not change caller's arguments in "KMeansPlusPlusClusterer":
A negative value for "maxIterations" will raise an exception.
o MATH-1629: Throw "ArithmeticException" instead of "MathArithmeticException".
o MATH-1630: Introduce "JdkMath" class for redirecting calls to either JDK or CM
implementations of the functions defined in "[Link]".
o Simplify assertions with simpler equivalent. Thanks to Arturo Bernal.
o MATH-1573: Remove redundant local variables. Thanks to Arturo Bernal.
o MATH-1572: Simplify conditional expressions. Thanks to Arturo Bernal.
o MATH-1616: Refactor "EmpiricalDistribution" class (in package
"[Link]").
o MATH-1595: Clean up "[Link]" package.
o MATH-1615: Remove class "StableRandomGenerator" (in package
"[Link]").
o MATH-1614: Refactor "SimplexOptimizer" (in package
"[Link]").
o MATH-1582: Transforms codes moved into a dedicated maven module.
o MATH-1578: ANN codes moved into a dedicated maven module.
o MATH-1575: Modularization set-up: All codes were moved into a "legacy" maven
module.
WIP: Functionalities should gradually be moved into dedicated modules.
o MATH-1562: Add "Automatic-Module-Name" entry to JAR manifest file. Thanks to
Frank Ulbricht.
o MATH-1554: Remove package "[Link]".
o MATH-1547: More flexible ranking of SOFM.
o MATH-1530: "SplineInterpolator": Improved performance. Thanks to Yassine
Damerdji.
o MATH-1524: Code reuse. Thanks to Chen Tao.
o MATH-1522: Remove generic parameter in clustering evaluator interfaces. Thanks
to Chen Tao.
o MATH-1523: Abstract class "ClusterEvaluator" replaced by an interface.
o MATH-1408: Do not use exceptions for control flow
o MATH-1362: Removed deprecated class "Incrementor" (replaced by
"[Link]").
o MATH-1506: Removed class "MultidimensionalCounter" (ported to "Commons
Numbers").
o MATH-1504: Removed class "BigFraction" (ported to "Commons Numbers").
o MATH-1446: Removed class "Fraction" (ported to "Commons Numbers").
o MATH-1469: Removed most codes in package "[Link]".
"Commons Math" now depends on "Commons Geometry".
o MATH-1481: "SobolSequenceGenerator": Update data file. Thanks to Théophile
Charmont.
o MATH-1389: Performance improvement for [Link]()
Thanks to Christoph Dibak.
o MATH-1377: "SimplexOptimizer": Slight performance improvement.
o MATH-1318: "[Link]": Improved performance (through the
use of
the reflection formula for negative arguments). Thanks to Eric Prescott-
Gagnon.
o MATH-1305: "AbstractRandomGenerator" and "BitsStreamGenerator": Slight
performance
improvement of the "nextBytes" method (particularly when the number of
requested bytes is a multiple of 4). Thanks to Rostislav Krasny.
o MATH-853: "MathRuntimeException" is now the base class for all commons-math
exceptions (except for "NullArgumentException" which extends
"NullPointerException").
o MATH-1205: The abstract class "AbstractStorelessUnivariateStatistic" does not
extend anymore from "AbstractUnivariateStatistic".
o MATH-1205: Default implementation of
"AbstractStorelessUnivariateStatistic#equals(Object)"
will only return true if both instances have the same type. Previously
different statistics were considered to be equal if their current state
happened to be equal.
o MATH-1205: Default implementations of
"AbstractStorelessUnivariateStatistic#evaluate(...)"
do not alter the internal state anymore. Instead a temporary copy of
the statistic is created for evaluation purposes.
o MATH-1206: Added new API methods in "[Link]" (package
"[Link]").
o MATH-1210: "QRDecomposition": include information about the condition that
triggers a "SingularMatrixException".
o MATH-757: Removed broken synchronization support in "ResizableDoubleArray".
o MATH-869: "SpearmansCorrelation" will now throw an
"MathIllegalArgumentException"
if provided with a "NaturalRanking" instance that uses "REMOVED" as
"NaNStrategy".
o MATH-869: "NullArgumentException" extends now "[Link]"
instead of "MathIllegalArgumentException".
o MATH-839: Renamed "cumulativeProbability(double, double)" to
"probability(double, double)"
in "IntegerDistribution" and "RealDistribution". Thanks to Gilles Sadowski.
o MATH-1155: WELL type pseudo-random number generators have been refactored:
the cached indirection index tables per instance are now stored
in a static instance. This reduces the initialization cost when
creating a new instance.
o MATH-825: Method "LaguerreSolver#laguerre(...)" has been made private. Thanks to
Gilles Sadowski.
o MATH-760: Field "windowSize" in "DescriptiveStatistics" has been made private.
Thanks to sebb.

Removed:
o MATH-1205: Removed methods "test(...)" from "AbstractUnivariateStatistic".
The already existing methods "MathArrays#verifyValues(...)" shall
be used instead.
o MATH-1050: Removed "ArithmeticUtils#pow(int, long)" and
"ArithmeticUtils#pow(long, long)".

For complete information on Apache Commons Math, including instructions on how to


submit bug reports,
patches, or suggestions for improvement, see the Apache Commons Math website:

[Link]

Common questions

Powered by AI

Apache Commons Math 4.0-beta1 addressed existing bugs and performance issues by incorporating numerous bug fixes and performance enhancements in its new components. Hundreds of issues were fixed, including exceptions such as "ArrayIndexOutOfBounds" in "MillerUpdatingRegression" and inaccurate methods in classes like "SimpsonIntegrator" and "EmpiricalDistribution". The update also included the implementation of faster and more accurate random number generators, improvement of the "MultivariateNormalDistribution" efficiency, and optimizations in methods like 'evaluate(...)' in "Variance" and performance improvements in "QRCodeComposition" .

The API for random number generation was significantly updated in Apache Commons Math 4.0-beta1 by introducing a new API under the package "o.a.c.m.rng", which aimed to improve performance and correctness. This update included faster 64-bit RNG implementations, which are approximately 35% faster than "java.util.Random" on 64-bit systems while offering better randomness quality. The update also involved refactoring the uniform random number generator functionality into a new API, creating an adaptor class to ensure compatibility with the old API, and addressing the 'seed problem' in its Javadoc. This comprehensive update aimed to modernize and streamline the generation of random numbers within applications .

The Apache Commons Math 4.0-beta1 release introduced several enhancements related to mathematical statistics to improve the library's functionality. Changes included adding "Percentile" support for weighted data and "AkimaSplineInterpolator" with alternative weight options, which improve data handling versatility. Additionally, classes dealing with statistical calculations, like "DescriptiveStatistics" which now allows "Double[]" as an initializer, offered more flexibility in data processing. These updates extended the library's ability to handle more statistical scenarios and improved performance with upgrades like enhanced "MultivariateNormalDistribution" which decreased overhead by removing unnecessary copying, allowing for larger scalability and more precise statistical computations .

Apache Commons Math 4.0-beta1 introduced an update to the "SimplexOptimizer" class, located in the package "o.a.c.m.legacy.optim.nonlinear.scalar.noderiv", to support simulated annealing. This integration allows the "SimplexOptimizer" algorithm, which is traditionally used for finding the minimum of univariate or multivariate functions, to apply a probabilistic technique to escape local minima by accepting worse solutions occasionally. This enhancement offers additional flexibility and robustness in optimization scenarios, effectively broadening the scope of the existing optimization algorithms to solve complex and non-linear problems .

The main motivations behind the structural changes in Apache Commons Math 4.0-beta1 were to address the complexity and bloat of the codebase by splitting lower-level functionality into more focused components such as Commons RNG, Commons Numbers, Commons Geometry, and Commons Statistics. This modularization was intended to enhance performance, improve API design, and implement bug fixes more effectively. Additionally, the changes aimed to manage the existing workload with limited human resources by creating more dedicated components to allow better maintenance and evolution of the library .

Apache Commons Math 4.0-beta1 introduced several new algorithms and interfaces. Notably, it added the "ElkanKMeansPlusPlusClusterer" and "MiniBatchKMeansClusterer" for advanced clustering tasks, the "Calinski-Harabsz" clustering evaluator for assessing clustering results, and an interface for ranking cluster lists. These innovations improved clustering capabilities by providing more efficient and reliable clustering methods. Furthermore, the "BicubicInterpolator" was enhanced to provide partial derivatives for more precise interpolation, crucial in mathematical modeling and scientific computations .

In the Apache Commons Math 4.0-beta1 release, legacy support and migration were managed by transitioning codes that had not undergone refactoring or were incomplete into 'legacy' modules. Users migrating to version 4.0 were advised to change the top-level package name from 'org.apache.commons.math3' to 'org.apache.commons.math4.legacy'. In terms of specific modular migrations, for instance, functionality previously covered by "CombinatoricsUtils" and "BigFraction" was moved to "Commons Numbers" and users were encouraged to transition to these dedicated modules to benefit from specialized maintenance and improvements .

The separation of the Apache Commons Math library into focused components, as seen in the 4.0-beta1 release, is expected to offer several advantages. It enhances modularity, allowing different components to evolve independently and potentially at different paces, improving maintainability. This separation can result in better-targeted optimizations and bug fixes within each component, thus increasing the overall performance. Additionally, it reduces the cognitive and resource burden required for developers to maintain and enhance these components due to smaller, more comprehensible codebases. This modular approach also encourages contributions from the community, offering more focused documentation and issue tracking .

Apache Commons Math 4.0-beta1 revised the handling of NaN values by using "Double.isNaN" instead of "x != x" in FastMath, which is a more standard and transparent method for NaN detection. This change reduces ambiguity and improves code readability, facilitating maintenance and debugging in numerical computations. Proper NaN handling ensures that computations return consistent and predictable results when faced with invalid or undefined numerical values, enhancing the robustness and reliability of the library’s function implementations .

In the Apache Commons Math 4.0-beta1 release, the class previously known as "FastMath" was refactored and renamed to "AccurateMath". Although detailed differences between the two are not explicitly mentioned in the release notes, the renaming implies a shift in focus from merely fast computations to improved accuracy, possibly incorporating more precise mathematical functions or algorithms. The "AccurateMath" class is now located in "org.apache.commons.math4.legacy.core.jdkmath". This change aligns with the strategy of enhancing the reliability and accuracy of numerical computations in the library .

You might also like