Roadmap for version 0.1

[psolymos]

Data input

• [x] read .xyz landmark data file
• [x] define data object (data is a list of length n, KxD matrices)
• [x] print method for data object (number of landmarks, dims, replicates; names, labels)
• [x] stack method to turn list into stacked X matrix
• [x] subset method: useful for bootstrap; subsetting landmarks, dimensions too

Nonparametric estimation

• [x] fitting function
• [x] return object structure
• [x] bootstrap method
• [x] print method
• [x] summary method
• [x] plot method: 2D
• [x] plot method: 3D (using FM based MDS and CIs)

Output from nonparametric estimation

• [x] compare 2 fitted models (numerator vs. denominator)
• [x] form matrix (option to sort the stacked version)
• [x] form difference matrix (FDM) with bootstrap based marginal CIs (option to sort the stacked version)
• [x] bootstrap based T statistic testing
• [x] growth matrix, growth difference matrix (GDM) with bootstrap based marginal CIs
• [x] plot for T statistic distribution
• [x] plot for FDM and GDM
• [ ] save outputs as sheets in xlsx file format

Parametric estimation

• [x] add different parametric forms forms, most basic: sigma^2*I
• [x] identifiability testing suite: check structural constraints
• [x] user provides a "structure" matrix as input, parametric fit and identifiability testing to follow

Documentation and testing

• [x] R help files for functions
• [x] imports, namespace, etc (R package checking stuff)
• [x] examples for the help files
• [x] tutorial describing the steps
• [x] add unit tests
• [x] add Linux/Windows CI/CD tests

Other

• [ ] explore other file formats (see how other packages do it, conversion methods)
• [ ] add parallel option to speed up bootstrap with multiple cores

For features outside of v 0.1, see https://github.com/psolymos/EDMAinR/issues/2

[psolymos]

Questions

• what to do for 2D samples? Have only XY columns, or add Z=0?

[psolymos]

3D plot:

• use FM (pairwise distances) with MDS,
• make a 3D plot with {rgl}
• color edges between landmarks according to >1 or <1 and CI based significance

[psolymos]

Parametric estimation of SigmaK:

• [x] SigmaK estimation + identifiability checks
• [x] use bootstrap results from object to estimate uncertainty for SigmaK
• [x] add a method to evaluate sensitivity to initial values

[psolymos]

Here is a more detailed roadmap for outstanding features and modifications:

Nonparametric estimation

• [x] remove mixed bootstrap, allow choosing reference (num, denom) for global T test
• [ ] find alternate naming (not T or F)

Visualization

• [x] find suitable divergent palette, option to change palette
• [x] implement 3D plot

Data checking

• [x] point cloud around landmarks (as in MorphoJ)
• [x] mean form + 1 individual at a time (indiv ID prnted), or use average?

Parametric estimation

• [x] check landmark names when working with pattern matrix
• [x] make non-estimability error optimal (i.e. allow exploration)
• [x] allow parameter names to be numeric (i.e. numbers as characters)

classification/clustering

• [x] use FM based log(T)-distances for HCA/MDS/k-means to see if there are groups
• [ ] predetermined groups, where does it belong? Wishart based likelihood distance ranking

Influential landmark

• [x] LOO type assessment, potentially leaving out multiple landmarks and assess global fit / CI?

GM/GDM revisions

• [x] GM/GDM revisions

Docs

• [x] add Rd pages
• [ ] updated tutorial (JSS style)