Bug in `PLNmixture()` when `clusters` is not a complete sequence

[mahendra-mariadassou]

PLNmixture() with backward (and potentially forward) smoothing fails when the clusters argument is not a complete sequence starting at 1.

library(PLNmodels)
#> This is packages 'PLNmodels' version 0.11.7-9720
#> Use future::plan(multicore/multisession) to speed up PLNPCA/PLNmixture/stability_selection.
data(trichoptera)
trichoptera <- prepare_data(trichoptera$Abundance, trichoptera$Covariate)
myMixtures <- PLNmixture(Abundance ~ 1 + offset(log(Offset)), clusters = 2:4, data = trichoptera)
#> 
#>  Initialization...
#> 
#>  Adjusting 3 PLN mixture models.
#>  number of cluster = 2   number of cluster = 3   number of cluster = 4 
#> 
#>  Smoothing PLN mixture models.
#>    Going backward +
#> Error in self$models[[k]]: indice hors limites

^{Created on 2023-02-04 with reprex v2.0.2}

The culprit are here and here as k refers both to the number of clusters and the position in the family of models.

[mahendra-mariadassou]

I realized while working on the fix that smoothing implicitly assumes that models in the family are adjacent (e.g. we have all models between 2 clusters and 10 clusters, not only the ones with 2, 5 and 10 clusters). I added a warning and easy fix but it might be useful to smooth with more spread out models. @jchiquet I'll work on a fix in a separate and let you decide whether it's worth including in the package.

[jchiquet]

Could you please an entry in NEWS.md?

thanks!

[mahendra-mariadassou]

Done. And you can have a look at branch mixture_smoothing to see how we can deal with incomplete sequences (e.g. c(1, 3, 5)).

• Pros: Allows for arbitrary sequences and return the expected number of  models (one per requested number of clusters)
• Cons: Requires estimating (but not always returning) a mixture model for every k in min(clusters):max(clusters). Wouldn't it be simpler to simply overwrite clusters with min(clusters):max(clusters) ?

[jchiquet]

This sounds good, and will match what the user has requested. Ok to merge in main/master for me. Do you keep all the model fitted in memory or just use them for the browsing during the slit/merge strategy ?

Thanks!

[mahendra-mariadassou]

I'll merge it. I keep only the requested models in memory, the others are only used during browsing.