Stdlib Distribution Functions

Help stdlib
1

Hi!
I’m relatively new to fortran only picking it up last year, but glad to be here looking deeper.

Reading some multiple-year old discussions and pr’s on the stdlib repo it seems there was a desire for expanding the existing 3 distribution functions in stdlib (normal, uniform and exponential) outwards to a wider range. Whilst at this point perhaps not up to date with current testing practices of stdlib layout there were 2 fairly complete looking pr’s for Gamma and Beta distribution functions.
Is there any greater desire for distribution functions such as those two, or perhaps Student’s t distributon or F distributon? Both are widely applicable in addition to Gamma and Beta and while not necessary within a stdlib it seems fairly appropriate.I’m iffy on their place within stdlib, what do other people think? I’ve not seen much discussion of this from my quick look around for quite some time.

1 Like
2

Hi @callum,

Welcome to the discourse, and thanks for engaging with and planning to contribute to stdlib!

Correct, there are PR286 and PR278 that are already fairly complete and just require some edits to bring them up to modern stdlib standards. I’ll happily review any PRs you submit on the stats side of stdlib and offer advice along the way.
Student’s t and F distributions functions would be very valuable, as they’d pave the way to the implementation of the most common statistical tests. Even if the tests themselves should be deemed outside the scope of stdlib, I think very common/basic distributions like t and F do have a place next to currently covered distributions (normal, exponential, and uniform).

The relatively modest stats modules have been touched on in @loiseaujc’s post, for example:

scipy.stats : stdlib provides some support for the uniform and Gaussian distribution, but it is nowhere as feature-complete as this scipy module.

I think any enhancement of these would be welcome by many.

2 Likes
3

My calc project has many distributions, each with random number generation (r), probability density (d), cumulative probability density (p), quantiles (q), estimation, and mean, standard deviation, skew, and kurtosis, modeled after R. I have looked at the results they give, but they should be tested more extensively against R and/or SciPy. Here are the implemented distributions.

Interpreter name family Statistical name
runif, dunif, punif, qunif Uniform distribution (default: Uniform(0,1))
rnorm, dnorm, pnorm, qnorm, fit_norm Normal (Gaussian) distribution
rexp, dexp, pexp, qexp, fit_exp, mssk_exp Exponential distribution
rgamma, dgamma, pgamma, qgamma, fit_gamma, mssk_gamma Gamma distribution
rlnorm, dlnorm, plnorm, qlnorm, fit_lnorm, mssk_lnorm Lognormal distribution
rt, dt, pt, qt, fit_t, mssk_t Student’s t distribution
rchisq, dchisq, pchisq, qchisq, fit_chisq, mssk_chisq Chi-square distribution
rf, df, pf, qf, fit_f, mssk_f F distribution (Fisher-Snedecor)
rbeta, dbeta, pbeta, qbeta, fit_beta, mssk_beta Beta distribution
rlogis, dlogis, plogis, qlogis, fit_logis, mssk_logis Logistic distribution
rsech, dsech, psech, qsech, fit_sech, mssk_sech Hyperbolic secant distribution
rlaplace, dlaplace, plaplace, qlaplace, fit_laplace, mssk_laplace Laplace (double exponential) distribution
rcauchy, dcauchy, pcauchy, qcauchy, fit_cauchy Cauchy distribution
rged, dged, pged, qged, fit_ged Generalized Error Distribution (GED, generalized normal / exponential power)
rhyperb, dhyperb, phyperb, qhyperb, fit_hyperb Hyperbolic distribution
1 Like
4

GitHub - urbanjost/M_datapac: The NIST DATAPAC package modularized and made available as an fpm(1) package might have a few additional cumulative distributions as well, although @Beliavsky’s list looks to be more extensive. The NIST DATAPLOT package contains a large amount of statistical routines in Fortran, DATAPAC being just one package extracted from it.

1 Like