Learners should be constructable without having their packages installed.
This is important, because otherwise the mlr_learners -> Dictionary conversion fails, when someone does not have the apcluster package installed.

Currently the naming of the labels is inconsistent. Some have clustering or learner appended and some don't.
Would also be nice to have some sort of labelling for the clustering typ:

• Deterministic (hard clustering):

  • Hierarchical classification methods: Agglomerative & divisive methods
  • Optimal partitions: k-means & variants

• Density-based methods: DBSCAN & variants

• Probabilistic (fuzzy clustering):

  • Model-based methods: Mixed distribution models
  • fuzzy k-means & variants

  This is done with properties, the man roxygen is not correct for this, some require updating.

Update the readme to be similar to https://github.com/mlr-org/mlr3torch or https://github.com/mlr-org/mlr3fda and use README.Rmd instead

Description

learner$predict(task) causes terminal to crash when eps is much lower than optimal. Tested in defult console, Radian and Posit Cloud.

Reproducible example

Steps to reproduce:

library(mlr3)
library(mlr3cluster)

task = mlr_tasks$get("usarrests")
learner = lrn("clust.dbscan", eps = 0.5, minPts = 5) # optimal eps between 30 and 40
learner$train(task)
preds = learner$predict(task = task)

Session info:

R version 4.2.2 (2022-10-31 ucrt)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 22000)

Matrix products: default

locale:
[1] LC_COLLATE=Portuguese_Brazil.1252  LC_CTYPE=Portuguese_Brazil.1252    LC_MONETARY=Portuguese_Brazil.1252 LC_NUMERIC=C
[5] LC_TIME=Portuguese_Brazil.1252

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] mlr3cluster_0.1.8 mlr3_0.15.0

loaded via a namespace (and not attached):
 [1] compiler_4.2.2       DEoptimR_1.0-11      mlr3misc_0.11.0      class_7.3-20         tools_4.2.2          prabclus_2.3-2       digest_0.6.31
 [8] mclust_6.0.0         uuid_1.1-0           jsonlite_1.8.4       checkmate_2.1.0      clue_0.3-64          lattice_0.20-45      rlang_1.1.0
[15] cli_3.6.1            parallel_4.2.2       cluster_2.1.4        globals_0.16.2       fpc_2.2-10           stats4_4.2.2         diptest_0.76-0
[22] grid_4.2.2           nnet_7.3-18          robustbase_0.95-0    data.table_1.14.8    listenv_0.9.0        R6_2.5.1             flexmix_2.3-19
[29] parallelly_1.35.0    kernlab_0.9-32       lgr_0.4.4            backports_1.4.1      codetools_0.2-18     modeltools_0.2-23    palmerpenguins_0.1.1
[36] MASS_7.3-58.1        future_1.32.0        paradox_0.11.1       crayon_1.5.2

I install the mlr3cluster through install.packages("mlr3cluster ") without error. However, the error occurs when I library it

Error: package or namespace load failed for ‘mlr3cluster’: .onLoad failed in loadNamespace() for 'mlr3cluster', details: call: rbindlist(l, use.names, fill, idcol) error: Item 2 has 6 columns, inconsistent with item 1 which has 7 columns. To fill missing columns use fill=TRUE.

I have tried to remove it and install again but of no use. Again, I tried to install for the source of github, and still does not work, and another error raise:
Error: package or namespace load failed for ‘mlr3cluster’ in get(Info[i, 1], envir = env): lazy-load database 'D:/R-4.1.0/library/mlr3cluster/R/mlr3cluster.rdb' is corrupt In addition: Warning message: In get(Info[i, 1], envir = env) : internal error -3 in R_decompress1
How could I fix it?

• https://cran.r-project.org/web/packages/clustMixType/clustMixType.pdf#page3
• https://tidyclust.tidymodels.org/reference/details_k_means_clustMixType.html

Getting the following warning in the tests:

Warning (test_mlr_learners_clust_mbatchkmeans.R:38:7): Learner properties are respected
`predict_MBatchKMeans()` was deprecated in ClusterR 1.3.0.
i Beginning from version 1.4.0, if the fuzzy parameter is TRUE the function 'predict_MBatchKMeans' will return only the probabilities, whereas currently it also returns the hard clusters

Consider adding Gaussian Mixture Models (GMM) with the option do do both hard and fuzzy clustering. See e.g., https://brilliant.org/wiki/gaussian-mixture-model/.

An alternative implementation is e.g. available in the clusteR package (https://cran.r-project.org/web/packages/ClusterR/vignettes/the_clusterR_package.html).

Having this algorithm in mlr3cluster would complete the set of fundamental clustering algorithm and thus, the need to rely on the routine of another package when applying GMM.

the last time the tests were run was 9 months ago
@damirpolat

Simple Example:

task <- tsk("usarrests")
kmeans_centers <- lapply(1:10, function(x) po("scale") %>>% lrn("clust.kmeans", centers = x))
design = benchmark_grid(
  tasks = task,
  learners = kmeans_centers,
  resamplings = rsmp("insample")
)
bmr = benchmark(design)
bmr$score(msr("clust.wss"))$clust.wss

will throw an output like

[1] 355807.82 114846.81 81862.19 79208.07 70152.06 68255.12 68148.43 63241.63 54304.11 43632.32

The output from wss is obviously too high to be scaled.

The problem can be found in MeasureClustInternal that takes the "raw" task without any preprocessing to calculate the features.
I think, this is probably only an issue that mlr3cluster suffers from, as all other Measures are only dependent on the predictions ...?

This could be avioded if there is any generic access to the preprocessed task in the pipeline.
In this case, one could exchange the taske in the function by the learner itself.
The problem is, if I enter the state of a trained pipeline, stored preprocessed Tasks are empty...

I am a bioinformatics PhD and I really appreciate your mlr3cluster package. This package provides many unsupervised clustering algorithms. However, I regret to find that the two most commonly used algorithms in bioinformatics analysis, consistency clustering and non-negative matrix factorization, are not included in this package. These two algorithms are widely used in the medical and biological fields. If these two algorithms are added to the package, the application scope will be greatly increased. I also hope to cite this package in my upcoming doctoral thesis. Thank you very much.

Many learners explicitly set the param set values to their defaults, e.g. in https://github.com/mlr-org/mlr3cluster/blob/master/R/LearnerClustKMeans.R#L46. This is not necessary. Also, every deviation from the default behavior should be documented on the learner's man page

After unloading mlr3cluster I think the dictionary items that were added should be removed again.
I made a similar PR here already: mlr-org/mlr3learners#258

Supported in the following package: https://github.com/mhahsler/stream check if predict method is supported. Also included in sklearn: https://scikit-learn.org/stable/modules/generated/sklearn.cluster.Birch.html#sklearn.cluster.Birch

Might as well incorporate BICO which is also written in C and there might be potential others as well that are implemented and not just serving as a wrapper such as dbscan.

in the .train() and .predict() method of the learner, task$data() is called twice, which should be avoided.

Prepare for release:

• git pull
• Check current CRAN check results
• Polish NEWS
• urlchecker::url_check()
• devtools::check(remote = TRUE, manual = TRUE)
• devtools::check_win_devel()
• rhub::check_for_cran()
• revdepcheck::revdep_check(num_workers = 4)
• Update cran-comments.md
• git push

Submit to CRAN:

• usethis::use_version('patch')
• devtools::submit_cran()
• Approve email

Wait for CRAN...

• Accepted 🎉
• git push
• usethis::use_github_release()
• usethis::use_dev_version()
• git push

Hey guys,

Whats wrong in this task? I can get it working with included task (e.g., task = mlr_tasks$get("usarrests") but the reprex below won't:

# metapackage
library(mlr3verse)

# task creation
task = TaskClust$new(
    id = "cars",
    backend = subset(
        mtcars,
        select = c(
            mpg,
            cyl,
            hp
        )
    )
)

# learner
learner = lrn("clust.kmeans")

# graph
graph = po("scale") %>>%
    po("learner", learner)

# convert graph to learner
glrn = as_learner(graph)
#> Error in .__GraphLearner__initialize(self = self, private = private, super = super, : 'graph' output type not 'Prediction' (or compatible with it)

SO question

Quite useful would be the implementation: flexclust::kcca with initcent = kmeanspp for k-means++ initialisation as referenced in Arthur, David, and Sergei Vassilvitskii. "k-means++: The advantages of careful seeding." In Soda, vol. 7, pp. 1027-1035. 2007.

• https://mlampros.github.io/ClusterR/reference/KMeans_rcpp.html
• https://github.com/tidymodels/tidyclust/blob/e6bd3420371677b41157ebb0883c30ab18d0a7b3/R/k_means.R#L172

See mlr-org/mlr3#732

• https://cran.r-project.org/web/packages/klaR/klaR.pdf
• https://tidyclust.tidymodels.org/reference/details_k_means_klaR.html

Currently there is no predict method for klaR::kmodes perhaps use tidyclust implementation: https://github.com/tidymodels/tidyclust/blob/e6bd3420371677b41157ebb0883c30ab18d0a7b3/R/predict_helpers.R#L28

R:

• https://tidyclust.tidymodels.org/index.html
• https://github.com/mhahsler/stream
• https://github.com/laperez/clustering

Python:

• https://pyclustering.github.io/docs/0.10.1/html/index.html
• https://scikit-learn.org/stable/modules/clustering.html