You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 46 Next »

Commons-Math Wish List

A list of feature requests made by Commons-Math users, contributors, and committers. These requests will be considered when deciding on new functionality for future releases. At any time, additional feature requests can be added simply by editing this page.

  • Add remedian statistic - [http://www.agoras.ua.ac.be/abstract/Remrob90.htm The Remedian: a Robust Averaging method for Large Data Sets]

  • Add alternative pseudo-random number generators (PRNGs)
    • "implement good random number generators, e.g., like those described in Numerical Recipes or Knuth". – AlChou

    • I highly recommend working with Paul Houle [http://www.honeylocust.com/RngPack/] to get these random number generators integrated into the math library. He has stated to me in the past that he is willing to relicense them under the Apache license. As well I beleive that with the BSD being compatable with Apache licensing, there would be little stopping us from integrating them ourselves as a derivative work. - Mark Diggory - Pluggable as of 1.1 release.

    • Investigate alternative methods for generating values from discrete distributions [http://www.jstatsoft.org/v11/i03/]

  • Add sparse matrix implementation. [http://mail-archives.apache.org/eyebrowse/BrowseList?listName=commons-dev@jakarta.apache.org&by=thread&from=736078]

  • Add Vector implementation (RealVector would extend RealMatrix)

  • Develop Genetic Algorithms Framework [http://markmail.org/message/jweowvj5hsy7omzc]

  • Resampling [http://markmail.org/message/u3diwc76m66r7qme]

  • Applied-mathematical/Mathematical-physics algorithms? - Henri Yandell
    • Examples, please? This item goes to the heart of what I consider an ongoing lack of consensus about what Commons-Math is supposed to be for. Should it include discipline-specific algorithms that do not overlap core numerical mathematical areas? Maybe, but we should discuss it. – AlChou

  • "Rolling" statistics with large windows but limited storage [http://mail-archives.apache.org/eyebrowse/ReadMsg?listName=commons-dev@jakarta.apache.org&msgNo=41870]

    • I don't understand; we already have some storage-less statistics. What am I missing from the post you refer to? – AlChou
    • Oh, I remember now that with the standard deviation algorithm I researched, in order to maintain a rolling window, it would have to store all the data in the window so it could delete the least recent one while adding the most recent one. Sorry for the denseness. – AlChou
  • Numerical Enhancements
    • Fast Fourier Transform - Code needs review and commit (BZ #36404)

    • [http://zxg32.blogchina.com/2787280.html Post-SOC TODO List] - Xiaogang Zhang

  • ANOVA in stat.inference package.

  • Multiple regression [http://mail-archives.apache.org/mod_mbox/jakarta-commons-dev/200401.mbox/%3c20040128222224.17393.qmail@web41704.mail.yahoo.com%3e]

  • QR decomposition (used for multiple regression)

  • Replace some of the discrete pmf and pdf methods with methods based on Catherine Loader's alogorithms [http://www.herine.net/stat/papers/dbinom.pdf]. These algorithms address some of the problems we experienced with large parameter values and have been adopted by many software packages including R.

  • Implement the Levenberg-Marquardt method [http://mathworld.wolfram.com/Levenberg-MarquardtMethod.html] – [CarlManaster]

  • Implement monte carlo simulation [http://en.wikipedia.org/wiki/Monte_Carlo_method]

  • [wiki:PrimeNumbers Prime Numbers Functionality] – [SharonLourduraj]

    • Naive Primality Testing
    • Probablistic Tests
    • Ofcourse, we will take it slowly, this area is vast and time consuming.
    • Implementing algorithms for practicality, and look into optimizing the algorithms (in terms of implementing it).
  • AbstractStorelessUnivariateStatistic.evaluate(...) and all the workhorse implementations in subclasses should be static methods. – NickGuenther
  • StandardDeviation has versions of .evaluate which take a precalculated mean. It would be nice if the same sort of thing could be had for all the other measures (e.g. skewness & kurtosis should be able to take both precalculated means and standard deviations) – NickGuenther
  • A 'between' function for checking number ranges. I've searched and could not find one, but if this functionality already exists, please contact me – AlexMarshall
  • No labels