This function performs clustering on mixed-type (continuous, discrete and categorical) longitudinal markers using Bayesian consensus clustering method with MCMC sampling
BCC.multi(
mydat,
id,
time,
center = 1,
num.cluster,
formula,
dist,
alpha.common = 0,
initials = NULL,
sigma.sq.e.common = 1,
hyper.par = list(delta = 1, a.star = 1, b.star = 1, aa0 = 0.001, bb0 = 0.001, cc0 =
0.001, ww0 = 0, vv0 = 1000, dd0 = 0.001, rr0 = 4, RR0 = 3),
c.ga.tunning = NULL,
c.theta.tunning = NULL,
adaptive.tunning = 0,
tunning.freq = 20,
initial.cluster.membership = "random",
input.initial.local.cluster.membership = NULL,
input.initial.global.cluster.membership = NULL,
seed.initial = 2080,
burn.in,
thin,
per,
max.iter
)Returns a BCC class model contains clustering information
list of R longitudinal features (i.e., with a length of R), where R is the number of features. The data should be prepared in a long-format (each row is one time point per individual).
a list (with a length of R) of vectors of the study id of individuals for each feature. Single value (i.e., a length of 1) is recycled if necessary
a list (with a length of R) of vectors of time (or age) at which the feature measurements are recorded
1: center the time variable before clustering, 0: no centering
number of clusters K
a list (with a length of R) of formula for each feature. Each formula is a twosided linear formula object describing both the fixed-effects and random effects part of the model, with the response (i.e., longitudinal feature) on the left of a ~ operator and the terms, separated by + operations, or the right. Random-effects terms are distinguished by vertical bars (|) separating expressions for design matrices from grouping factors. See formula argument from the lme4 package
a character vector (with a length of R) that determines the distribution for each feature. Possible values are "gaussian" for a continuous feature, "poisson" for a discrete feature (e.g., count data) using a log link and "binomial" for a dichotomous feature (0/1) using a logit link. Single value (i.e., a length of 1) is recycled if necessary
1 - common alpha, 0 - separate alphas for each outcome
List of initials for: zz, zz.local ga, sigma.sq.u, sigma.sq.e, Default is NULL
1 - estimate common residual variance across all groups, 0 - estimate distinct residual variance, default is 1
hyper-parameters of the prior distributions for the model parameters. The default hyper-parameters values will result in weakly informative prior distributions.
tuning parameter for MH algorithm (fixed effect parameters), each parameter corresponds to an outcome/marker, default value equals NULL
tuning parameter for MH algorithm (random effect), each parameter corresponds to an outcome/marker, default value equals NULL
adaptive tuning parameters, 1 - yes, 0 - no, default is 1
tuning frequency, default is 20
"mixAK" or "random" or "PAM" or "input" - input initial cluster membership for local clustering, default is "random"
if use "input", option input.initial.cluster.membership must not be empty, default is NULL
input initial cluster membership for global clustering default is NULL
seed for initial clustering (for initial.cluster.membership = "mixAK") default is 2080
the number of samples disgarded. This value must be smaller than max.iter.
the number of thinning. For example, if thin = 10, then the MCMC chain will keep one sample every 10 iterations
specify how often the MCMC chain will print the iteration number
the number of MCMC iterations.
# import dataframe
data(epil)
# example only, larger number of iteration required for accurate result
fit.BCC <- BCC.multi (
mydat = list(epil$anxiety_scale,epil$depress_scale),
dist = c("gaussian"),
id = list(epil$id),
time = list(epil$time),
formula =list(y ~ time + (1|id)),
num.cluster = 2,
burn.in = 3,
thin = 1,
per =1,
max.iter = 8)
Run the code above in your browser using DataLab