commons-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Mark R. Diggory" <>
Subject Re: [MATH] How stable are the matrix classes?
Date Mon, 09 Aug 2004 12:07:21 GMT
Hi Kim,

I think we have a bit of work to complete in extending the api into 
areas of multivariate analysis/statistics and would be interested in 
implementations in that area. I think the big thing is that we would 
like to see implementation be based on the existing library architecture 
as well, so goal might be to see if these could be easily modified to 
work with the existing Matrix/Statistics classes in our api.


Kim van der Linde wrote:
> Hi Phil,
> Thanks for the answer, and I was talking about numerical stability.
> On an other track, if the math group is interested, I can provide some
> classes for Multivariate Minimum Volume Ellipsoid outlier detection,
> Covariances, common matrix types as SSCP, covariance and correlation,
> multivariate euclidian and mahalanobis distances and maybe some other
> classes as model 2 regressions (RMA: Reduced Major Axis). Interested?
> Kim
> Phil Steitz wrote:
>> Kim van der Linde wrote:
>>> Hi All,
>>> I have a question. How stable are the matrix classes as implemented?
>>> Cheers,
>>> Kim
>> Hi Kim,
>> If your question is about the API, then the answer is that we are 
>> planning no changes prior to the imminent 1.0 release.  If your 
>> question is about numerical stability, performance or correctness 
>> there are two things to say:
>> 1) The javadoc describes the algorithms used to perform matrix 
>> operations.  The algorithms are general purpose, so they will not 
>> always give the best results (or performance) for all matrices. For 
>> most practical problems, the implementations should work fine. Have a 
>> look at the docs and consult a numerical linear algebra text (or a 
>> numerical analyst) or ask more specific questions here if you want to 
>> know about individual operations. Eventually, we will provide support 
>> for a wider variety of algorithms.  For 1.0, what you see now is what 
>> you get.
>> 2) Our confidence in implementation correctness is based pretty much 
>> entirely on the unit tests at this point. This is new code, not yet 
>> released. We are in the process of cutting a release candidate 
>> including these classes. User feedback and/or additional test cases 
>> will be greatly appreciated.
>> Phil
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail:
>> For additional commands, e-mail:

Mark Diggory
Software Developer
Harvard MIT Data Center

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message