TODO.md

In TrN000/norMmix: Direct MLE for Multivariate Normal Mixture Distributions

1. ssClara2kL() {was 'ssClaraL'}: seems to give too small samples in several of the BSc thesis simulations. Provide "better" alternative!

2. Provide as.norMmix() generic and as.norMmix.nor1mix() to get 1-D comparisons and MW from CRAN package nor1mix

3. norMmix() has trouble with 1 dimensional and/or 1 component mixtures. deciding criteria needed.

4. documentation need examples and arguments of plot.... are not described. OUT OF DATE:Documentation for plot methods now completely broken. Arguments like newWindow no longer supported. Needs to be rewritten for 2d and >2d.

5. Currently always use full covariance parameter Sigma, even in cases such as EII etc. Should we allow both (full and minimal) parametrizations

6. manyMLE() is not yet exported & documented; before doing so, needs tweaks (by MM):

   • savdir and name with default name to save as "*rds": nice idea but should not be part of manyMLE but separate small utility.
   • modify argument checking; use namespace global norModels <- eval(formals(norMmix)$model)

7. BUG in either npar() or nMm2par() -- see 'FIXME' in man/nMm2par.Rd

8. Start a inst/NEWS.md file.

9. finish regression test in test-norMmix.R. Should test construction of Sigma from various inputs.

============================================================================== DONE:

1. Also port most of the "fit.R" from norMmix_Bthesis; i.e., the fitnMm() function {with better name!} --> ~/Betreute-Arbeiten/NicolasTrutmann/BSc_thesis+MM/norMixBthesis/R/fit.R and then all the <fun>.fittednorMmix() methods {but probably also change the name of the class from "fittednorMmix" to something like "manyNormixMLE" --> Nicolas Trutmann has done it (Jun 21 2020): name manyMLE() ==> ~/R/D/GH/norMmix/R/fit.R

2. norMmix.Rd does not document use of non-array covar. mats. as init. values

3. decide on par args in ndplot and how to leave it exposed to the user. maybe do.call(par, parargs) construct?? and put parargs=NULL in arguments. user can then overwrite first call to par().

==> MM: Using sfsmisc::mult.fig() and allow the user to change defaults for mult.fig() is easier.

===> MUCH better is really to change the setup completely, use graphics :: pairs.default(.) by providing a correct panel = function(.) !!!

1. --> R/plot.R: FIXME: plot2d() <--> plotnd() are NOT compatible in their defaults

2. rename nMmcol to Trubetskoy10

3. document or erase mention of parcond.

4. norMmix(...) creation does NOT check Sigma in case of restricted parametrizations, but just assumes the caller of norMmix() makes no mistake. This is clearly too optimistic! -> split into "private" and public method, where the public method does argument checking and private one assumes correct args. Should use the private one in MLE algorithm as it gets called there in a loop.

TrN000/norMmix documentation built on Sept. 9, 2024, 4:20 a.m.