Simple C Statistics

[Ross A. Finlayson]

Hi,

I wonder if there is a simple library or set of functions for basic
statistical functions, like what may be on a pocket calculator, without
getting into linking into the language runtimes of the S or R
languages.

For example I just want mean, median, mode, variance, and stuff, and
then the stuff that follows from that. Maybe, something midrange
between (x+y) / 2 and SAS/Spss/... statlib? libstat? stat lib stat,
stat, lib?

Actually I just want to look at it, not probably actually use it, but I
might implement some of its notions after seeing it, especially if
they're imperative instead of functional.

Actually this is just a troll and off-topic for comp.lang.c. While
that is so, I guess I could ask on sci.stat.math, sincerely.

I'll tell you why, I'm trying to analyze digital data with noise in it,
but I just want to flip to correctness about 1/100 of the bad bits, not
analyze its entropy, because there's not a carrier signal, and I'm
ignorant about signal processing. It's not even that hard a problem,
it's mostly elastic knapsack, and the weather.

Anyways I'm interested in seeing some examples of C implementations of
statistical methods. I hope you might be able to refer me to some.
Ah... now I have typed "statistics" into a search engine for
comp.lang.c.

Thank you,

Ross F.

 

[beliavsky]

Hi,

I wonder if there is a simple library or set of functions for basic
statistical functions, like what may be on a pocket calculator, without
getting into linking into the language runtimes of the S or R
languages.

The GNU Scientific Library has C code for statistics, see
http://www.gnu.org/software/gsl/manual/gsl-ref_toc.html , as does the
commercial Numerical Recipes in C, see http://www.nr.com . Gretl at
http://gretl.sourceforge.net/ is an open source statistical package
written in C . Statcodes at http://www.astro.psu.edu/statcodes/ has
links to public domain Fortran and C code for statistics.

 

[Ross A. Finlayson]

The GNU Scientific Library has C code for statistics, see
http://www.gnu.org/software/gsl/manual/gsl-ref_toc.html , as does the
commercial Numerical Recipes in C, see http://www.nr.com . Gretl at
http://gretl.sourceforge.net/ is an open source statistical package
written in C . Statcodes at http://www.astro.psu.edu/statcodes/ has
links to public domain Fortran and C code for statistics.

Hi Beliavsky,

Thank you. That is mind-boggling.

Basically I figure there is a sample, the sample space, the population,
which might be the same thing as the sample space, the probability of a
sample being selected at "random" for various distributions of metrics
of the samples of the population, which might be unknown or
parameterized as opposed to knowing every element of the sample space
in advance in which case it is fixed, the probability, P, of the binary
predicate, into fuzzy and multivalent metrical predicates, these are
words I string together without qualification.

That's a problem in "statistics", those web pages appear immaculately
respectable, but the abuse in statistics is of the metrics and sample
spaces from the actual population.

typedef void* sample_t;

I'm way back on regression, ANOVA, ANCOVA, MANOVA, MANCOVA,
curve-fitting, best-guess estimates for algorithm selection on
instrumented algorithm conditions and contraints, and other nonsense
drooled onto the keyboard, I only had three or four years of college
stat.

That's about statistics, about mathematics and logic I'm an absolutist.

I think the Bayesian methods are very useful for decision making based
on essentially neural networks, as neural networks are modelled in some
sense Bayesianly.

The neural net (keywords forward, backward, and mixed mode propagation,
transfer and decision function, Kohenen, Hopfield, Markov, binary and
analog weights, graph, network and circuit theories, which are the same
thing) is also a very useful tool, and, the implementations are not
necessarily minimal but they are not necessarily massive.

So, my next question is "what C libraries are available for portable
neural net implementations", and an answer may well be "go find it
yourself."

http://www.google.com/search?q=Bayesianly+neural

That is mildly disturbing.

Warm regards,

Ross F.