ordinations

[antagomir]

I noticed that the ordination methods (for PCA, MDS/PCoA, NMDS, UMAP at least) could be more standardized, with more clear examples in OMA how to use.

For instance (this mixes mia, miaViz & OMA issues but these are interrelated).

1. It may now depend on the method how feature importances (per each axis) are retrieved; in PCA it is the loadings, in nonlinear methods it could be feature-axis correlations unless other methods are available. There could be a standard method to retrieve loadings for any standard ordination; or at least a slot with standard name for this in the metadata. We previously thought that this is needed but it might help to standardize workflows across different ordinations, and provide default methods for the importance. E.g. something like

tse <- addPCA(tse, ...); 
x <- getFeatureImportance(tse, "PCA"); 
plotLoadings(x) or plotLoadings(tse, "PCA") 

2. More clear examples in OMA how to retrieve and visualize feature importances (names can be confusing, now terms rotation (attribute in reducedDim), loadings (in plotLoadings), and importances/weights (in some ML literature) are all in use). The OMA chapter that demonstrates these could be revised for clarity.

3. Function to add explained axis variances might also be useful. At least for PCA. For nonlinear methods there may be more suitable methods (e.g. stress for MDS).