natto is a wild package, but it has some house-trained functions that would not put us all ashamed if published on CRAN. I am considering to have a release git branch of those functions and publish them on CRAN. That package could be called natto with another (lower) version number than the wild one, or it could be a petty-bourgeois variant of natto such as tofu (both of these names seem to be available on CRAN at the moment).
The functions that should be published are:
canneddist: canned dissimilarity indices for vegan::designdist.distconstrain: the core of vegan::dbrda.gradrescale: rescaling of decorana applied to observed gradient.humpfit: first in vegan, but transferred here, and would make a return to CRANiap: basically number of companion species for each species with nullmodel tests. The name comes from lichen bioindication where this was used as an Index of Atmospheric Purity.infoclust: this is the classical Lance & Williams information clustering.polarord: the classic Bray & Curtis polar ordination with selection of endpoints (and probably similar to PC-Ord). Useful in education, though some people still seem to use this.posvectord & spvectord: both are Orlóci classic, position vector ordination and a kind of PC using species (variables) as surrogate PC axes. Useful in education.qrao: Rao's quadratic entropy and Rao's dissimilarities with phylogenies, taxonomies etc.raostand: standardization of the data so that Euclidean distances are Rao's distances, Simpson diversities Rao's quadratic entropies, and any other method using the data will be based on phylogenies, taxonomies etc.rdecorana: Implementation of Fortran decorana in R allowing easier inspection of the method and fiddling with code. Useful in education and tinkering, but real analysis naturally should use faster Fortran code in vegan.taxondist: Clarke's taxonomic dissimilarities that were requested for vegan but got here, and could get to CRAN.
Functions that could get published, but I'm not sure at all:
distMeans to find dissimilarities to the centroid. Currently this would fail in CRAN tests since it uses non-exported vegan functions with :::. Would require changes in vegan exports.diverclust & diverdist: cluster analysis based on Rényi diversity. Walking close to the wild side. Starts well, but may suffer from "chaining": once a cluster becomes very diverse, it can absorb anything.ginkgogram: fails in CRAN tests because accesses directly non-exported R functions when modifying stats:::plot.hclust.peelhull & peelellipse: I see methodological weaknesses (and similar has been published in other packages). If published, could be plugged to vegan::ordihull, vegan::ordiellipse to draw convex hulls or ellipses that enclose a specified proportion of points but ignore outliers.
Functions that should not be published:
- walking on the wild side:
bayesjaccard with friends, cNMDS, gnlmod, respthresh, sdca
- things I don't trust:
mrankdist, recomm, resassoc, slopediff
- things too special or adding dependencies:
as.dist.deldir, as.prcomp.rda, subdiag
Functions that could be moved from vegan:
MOStest- Perhaps many others, but that came first to my mind (has annoyed me a long time).
natto is a wild package, but it has some house-trained functions that would not put us all ashamed if published on CRAN. I am considering to have a release git branch of those functions and publish them on CRAN. That package could be called natto with another (lower) version number than the wild one, or it could be a petty-bourgeois variant of natto such as tofu (both of these names seem to be available on CRAN at the moment).
The functions that should be published are:
canneddist: canned dissimilarity indices forvegan::designdist.distconstrain: the core ofvegan::dbrda.gradrescale: rescaling ofdecoranaapplied to observed gradient.humpfit: first in vegan, but transferred here, and would make a return to CRANiap: basically number of companion species for each species with nullmodel tests. The name comes from lichen bioindication where this was used as an Index of Atmospheric Purity.infoclust: this is the classical Lance & Williams information clustering.polarord: the classic Bray & Curtis polar ordination with selection of endpoints (and probably similar to PC-Ord). Useful in education, though some people still seem to use this.posvectord&spvectord: both are Orlóci classic, position vector ordination and a kind of PC using species (variables) as surrogate PC axes. Useful in education.qrao: Rao's quadratic entropy and Rao's dissimilarities with phylogenies, taxonomies etc.raostand: standardization of the data so that Euclidean distances are Rao's distances, Simpson diversities Rao's quadratic entropies, and any other method using the data will be based on phylogenies, taxonomies etc.rdecorana: Implementation of Fortrandecoranain R allowing easier inspection of the method and fiddling with code. Useful in education and tinkering, but real analysis naturally should use faster Fortran code in vegan.taxondist: Clarke's taxonomic dissimilarities that were requested for vegan but got here, and could get to CRAN.Functions that could get published, but I'm not sure at all:
distMeansto find dissimilarities to the centroid. Currently this would fail in CRAN tests since it uses non-exported vegan functions with:::. Would require changes in vegan exports.diverclust&diverdist: cluster analysis based on Rényi diversity. Walking close to the wild side. Starts well, but may suffer from "chaining": once a cluster becomes very diverse, it can absorb anything.ginkgogram: fails in CRAN tests because accesses directly non-exported R functions when modifyingstats:::plot.hclust.peelhull&peelellipse: I see methodological weaknesses (and similar has been published in other packages). If published, could be plugged tovegan::ordihull,vegan::ordiellipseto draw convex hulls or ellipses that enclose a specified proportion of points but ignore outliers.Functions that should not be published:
bayesjaccardwith friends,cNMDS,gnlmod,respthresh,sdcamrankdist,recomm,resassoc,slopediffas.dist.deldir,as.prcomp.rda,subdiagFunctions that could be moved from vegan:
MOStest