Metavisualizations using vegan

[csapou]

I would like a feature in vegan to compare different data processing methods visually.

I have worked on this extensively to create an R package https://github.com/csapou/metaviswiz, but have also realized that it is basically a wrapper around vegan functions and a few other compositional approaches. I have also tried to publish this with no luck (https://www.biorxiv.org/content/10.1101/859157v1), however, not due to the approach.

The basic concept is
To create a list of different distance/dissimilarity matrices ➡️ perform a procrustes analysis pairwise between all the distance/dissimilarity matrices and bind 1-t0 (the Procrustes correlation) into a distance matrix ➡️ visualize the distance matrix

I also wanted to implement other features such as a density plot of the correlations and validation plots of the generated visualization of the distance matrix comparing processing methods.

I'm still motivated to work on it and believes it is a really cool feature, but I don't know how to implement it in vegan. Really hope to get some feedback on how this could be achieved.

[jarioksa]

This all sounds very exciting. However, I think that this kind of visualisations are are better suited for a separate package building upon vegan. We already have some such packages: BiodiversityR (in CRAN) provides a GUI for several vegan functions and also adds some new alternative methods, ggvegan (in GitHub) provides an interface to ggplot2 graphics, and vegan3d (in CRAN) provides an interface to dynamic 3D graphics with RGL (and also static 3D graphics with scatterplot3d). Adding such a metavisualisation package would be a good alternative.

I can help in janitorial job in package development if you are uncertain how to proceed. However, there are several howto's and other instructions floating around on the web, and if you are an R-Studio user, it has dedicated tools for building package infrastructure.

You may also consider issue #108 for new visualisations.

[csapou]

I see your point, I did not think about all the dependencies it would add. I'm not that familiar with ggvegan and vegan3d, since I always did my own ggplots extracting the information from the vegan objects. Looking into their documentation, I would guess my sensitivity analysis is beyond their scope? As I mentioned I'm still really motivated to work further on it, also if this means making my own package (added bonus). I got stuck due to several reasons:
-Is it relevant? I personally think it is really interesting if data is robust to different types of "reasonable" processing. But I also really don't want this tool to be used for picking the analyses confirming a hypothesis.
-Understanding the fundamentals of all the function parameters in decostand and vegdist. I looked through the source code, but since it is written in C+ (maybe another language?) I had a hard time to understand them all. Furthermore, sometimes combining the two is nonsensical or have it own distance/dissimilarity name.
-My own abilities. All the functions I had envisioned are working, but with a limited number of methods. However, I'm sure it can be made much better. I also don't have any experience with creating packages.