compare_mmm() fits a mixed Markov model with build_mmm() for each
number of subgroups in k and returns one row per fit with its
log-likelihood, information criteria (AIC, BIC, ICL), overall average
posterior probability (AvePP), entropy and convergence. Every fit uses
the same data and the same build_mmm() settings.
Lower AIC, BIC and ICL indicate more support relative to the other fits.
ICL adds a penalty for uncertain classification to BIC, so it favours
subgroups that separate cleanly (Biernacki, Celeux & Govaert, 2000);
AvePP and entropy describe that separation within each fit. The criteria
do not settle k on their own: check that every fit converged, that the
subgroups' transition matrices of the chosen fit differ in interpretable
ways (summary() of that fit prints them), and, when the criteria keep
falling at the largest k, extend the range.
Usage
compare_mmm(
data,
k = 2:5,
return_fits = FALSE,
n_starts = 50L,
max_iter = 200L,
tol = 1e-06,
smooth = 0.01,
seed = NULL,
covariates = NULL,
covariate_effect = c("em", "posthoc"),
estimator = c("auto", "firth", "multinom", "chisq")
)
# S3 method for class 'mmm_compare'
print(x, ...)
# S3 method for class 'mmm_compare'
summary(object, ...)
# S3 method for class 'mmm_compare'
plot(x, ...)Arguments
- data
The sequences, in any form
build_mmm()accepts: a wide data frame, anetobjectorcograph_networkbuilt from sequences, or atnamodel.- k
The numbers of subgroups to fit, a non-empty vector of whole numbers of at least 2; one model is fitted per value. Default
2:5.- return_fits
Whether to keep the fitted models. With
TRUE, thenet_mmmfits are attached to the result as its"fits"attribute, a list named byk, so the chosen model needs no second run of the EM. DefaultFALSE, which returns only the comparison table.- n_starts
How many starting configurations to try, a positive whole number. More starts make it likelier that the best local maximum is found. Default
50.- max_iter
The most EM iterations a start may run, a positive whole number. Default
200.- tol
The convergence tolerance: EM stops when the log-likelihood changes by less than this between iterations (with
covariate_effect = "em", also when its relative change does). A finite positive number; default1e-6.- smooth
A pseudo-count added to every transition count before the matrices are estimated, so a transition unseen in a subgroup gets a small probability instead of zero. A finite non-negative number; default
0.01.- seed
A seed for reproducible starts, applied afresh to each fit (every
kstarts from the same random state). DefaultNULL.- covariates
Variables describing each sequence (such as a course or a grade), given as a column name, a character vector of names, a one-sided formula or a data frame, as in
build_clusters(). For anetobject, names are looked up in its metadata first, so a typical call isbuild_mmm(net, k = 3, covariates = "session_label"). How they enter the model is set bycovariate_effect. DefaultNULL.- covariate_effect
How
covariatesenter the model."em"(default) makes the mixing proportions depend on them during EM, so they shape the subgroups found and add(k - 1) * pparameters forpcovariate columns; rows with a missing covariate are dropped first, with a warning."posthoc"fits a plain mixture on every sequence and relates the covariates to the subgroups only afterwards, with a multinomial logit, so neither their values nor their missingness change the subgroups. Ignored withoutcovariates.- estimator
How the after-fit multinomial logit is estimated; it never affects the EM.
"auto"(default) uses Firth's penalised likelihood (brglm2::brmultinom(), finite under separation) when a cell of the subgroup-by-covariate table has fewer than 5 sequences, andnnet::multinom()otherwise."firth"and"multinom"force one of them ("multinom"warns about separation risk), and"chisq"runs descriptive tests instead of a logit.build_clusters()has the details. Ignored withoutcovariates.- x
- ...
For the
print(),summary()andplot()methods: not used; any extra argument raises an error.- object
For the
summary()method: anmmm_comparetable.
Value
An mmm_compare object, a data frame with one row per value of
k. print() shows the table with a best
column marking the lowest BIC ("<-- BIC") and, when it falls on a
different row, the lowest ICL ("<-- ICL"), and returns that marked
table invisibly. summary() returns the same table as a plain data
frame with a best column holding "BIC", "ICL" or ""; when one
row has both minima it is marked "BIC" only. plot() draws BIC, AIC
and ICL against k, and returns the ggplot invisibly after printing
it. The table's columns are k, log_likelihood, AIC, BIC,
ICL, AvePP, Entropy and converged; return_fits = TRUE adds the
"fits" attribute.
References
Biernacki, C., Celeux, G., & Govaert, G. (2000). Assessing a mixture model for clustering with the integrated completed likelihood. IEEE Transactions on Pattern Analysis and Machine Intelligence, 22(7), 719–725. doi:10.1109/34.865189
Examples
seqs <- data.frame(V1 = sample(c("A","B","C"), 30, TRUE),
V2 = sample(c("A","B","C"), 30, TRUE))
comp <- compare_mmm(seqs, k = 2:3, n_starts = 1, max_iter = 10, seed = 1)
comp
#> MMM Model Comparison
#>
#> k log_likelihood AIC BIC ICL AvePP Entropy converged
#> 2 -62.32972 158.6594 182.4798 184.6518 0.9647836 0.2121121 TRUE
#> 3 -62.32987 176.6597 213.0909 217.2174 0.9343265 0.2487206 FALSE
#> best
#> <-- BIC
#>
# \donttest{
seqs <- data.frame(
V1 = sample(LETTERS[1:3], 30, TRUE), V2 = sample(LETTERS[1:3], 30, TRUE),
V3 = sample(LETTERS[1:3], 30, TRUE), V4 = sample(LETTERS[1:3], 30, TRUE)
)
comp <- compare_mmm(seqs, k = 2:3, seed = 42)
print(comp)
#> MMM Model Comparison
#>
#> k log_likelihood AIC BIC ICL AvePP Entropy converged
#> 2 -120.8229 275.6458 299.4661 299.9897 0.9920817 0.03537724 TRUE
#> 3 -115.1023 282.2045 318.6357 319.2157 0.9908021 0.03304310 TRUE
#> best
#> <-- BIC
#>
# Retain the fits so the chosen model needs no re-run; summary() marks
# the minimum-BIC and minimum-ICL rows in its `best` column.
comp_with_fits <- compare_mmm(seqs, k = 2:3, seed = 42, return_fits = TRUE)
summary(comp_with_fits)
#> k log_likelihood AIC BIC ICL AvePP Entropy converged
#> 1 2 -120.8229 275.6458 299.4661 299.9897 0.9920817 0.03537724 TRUE
#> 2 3 -115.1023 282.2045 318.6357 319.2157 0.9908021 0.03304310 TRUE
#> best
#> 1 BIC
#> 2
# }