sample_fiSAN_sparsemix {SANple} | R Documentation |
Sample fiSAN with sparse mixtures
Description
sample_fiSAN_sparsemix
is used to perform posterior inference under the finite-infinite shared atoms nested (fiSAN) model with Gaussian likelihood.
The model uses a Dirichlet process mixture prior at the distributional level,
and a sparse (overfitted) Dirichlet mixture (Malsiner-Walli et al., 2016) at the observational level.
The algorithm for the nonparametric component is based on the slice sampler for DPM of Kalli, Griffin and Walker (2011).
Usage
sample_fiSAN_sparsemix(nrep, burn, y, group,
maxK = 50, maxL = 50,
m0 = 0, tau0 = 0.1, lambda0 = 3, gamma0 = 2,
hyp_alpha1 = 1, hyp_alpha2 = 1,
hyp_beta = 10,
eps_beta = NULL,
alpha = NULL, beta = NULL,
warmstart = TRUE, nclus_start = NULL,
mu_start = NULL, sigma2_start = NULL,
M_start = NULL, S_start = NULL,
alpha_start = NULL, beta_start = NULL,
progress = TRUE, seed = NULL)
Arguments
nrep |
Number of MCMC iterations. |
burn |
Number of discarded iterations. |
y |
Vector of observations. |
group |
Vector of the same length of y indicating the group membership (numeric). |
maxK |
Maximum number of distributional clusters |
maxL |
Maximum number of observational clusters |
m0 , tau0 , lambda0 , gamma0 |
Hyperparameters on |
hyp_alpha1 , hyp_alpha2 |
If a random |
hyp_beta , eps_beta |
If a random |
alpha |
Distributional DP parameter if fixed (optional). The distribution is |
beta |
Observational Dirichlet parameter if fixed (optional). The distribution is Dirichlet( |
warmstart , nclus_start |
Initialization of the observational clustering.
|
mu_start , sigma2_start , M_start , S_start , alpha_start , beta_start |
Starting points of the MCMC chains (optional). Default is |
progress |
show a progress bar? (logical, default TRUE). |
seed |
set a fixed seed. |
Details
Data structure
The finite-infinite common atoms mixture model is used to perform inference in nested settings, where the data are organized into groups.
The data should be continuous observations
, where each
contains the
observations from group
, for
.
The function takes as input the data as a numeric vector
y
in this concatenated form. Hence y
should be a vector of length
. The
group
parameter is a numeric vector of the same size as y
indicating the group membership for each
individual observation.
Notice that with this specification the observations in the same group need not be contiguous as long as the correspondence between the variables
y
and group
is maintained.
Model
The data are modeled using a univariate Gaussian likelihood, where both the mean and the variance are observational-cluster-specific, i.e.,
where is the observational cluster indicator of observation
in group
.
The prior on the model parameters is a Normal-Inverse-Gamma distribution
,
i.e.,
,
(shape, rate).
Clustering
The model performs a clustering of both observations and groups.
The clustering of groups (distributional clustering) is provided by the allocation variables , with
The distribution of the probabilities is ,
where GEM is the Griffiths-Engen-McCloskey distribution of parameter
,
which characterizes the stick-breaking construction of the DP (Sethuraman, 1994).
The clustering of observations (observational clustering) is provided by the allocation variables , with
The distribution of the probabilities is for all
.
Value
sample_fiSAN_sparsemix
returns four objects:
-
model
: name of the fitted model. -
params
: list containing the data and the parameters used in the simulation. Details below. -
sim
: list containing the simulated values (MCMC chains). Details below. -
time
: total computation time.
Data and parameters:
params
is a list with the following components:
nrep
Number of MCMC iterations.
y, group
Data and group vectors.
maxK, maxL
Maximum number of distributional and observational clusters.
m0, tau0, lambda0, gamma0
Model hyperparameters.
- (
hyp_alpha1,hyp_alpha2
) oralpha
Either the hyperparameters on
(if
random), or the value for
(if fixed).
- (
hyp_beta,eps_beta
) orbeta
Either the hyperparameter on
and MH step size (if
random), or the value for
(if fixed).
Simulated values:
sim
is a list with the following components:
mu
Matrix of size (
nrep
,maxL
). Each row is a posterior sample of the mean parameter for each observational cluster.
sigma2
Matrix of size (
nrep
,maxL
). Each row is a posterior sample of the variance parameter for each observational cluster.
obs_cluster
Matrix of size (
nrep
, n), with n =length(y)
. Each row is a posterior sample of the observational cluster allocation variables.
distr_cluster
Matrix of size (
nrep
, J), with J =length(unique(group))
. Each row is a posterior sample of the distributional cluster allocation variables.
pi
Matrix of size (
nrep
,maxK
). Each row is a posterior sample of the distributional cluster probabilities.
omega
3-d array of size (
maxL
,maxK
,nrep
). Each slice is a posterior sample of the observational cluster probabilities. In each slice, each columnis a vector (of length
maxL
) observational cluster probabilitiesfor distributional cluster
.
alpha
Vector of length
nrep
of posterior samples of the parameter.
beta
Vector of length
nrep
of posterior samples of the parameter.
maxK
Vector of length
nrep
of the number of distributional DP components used by the slice sampler.
References
Kalli, M., Griffin, J.E., and Walker, S.G. (2011). Slice Sampling Mixture Models, Statistics and Computing, 21, 93–105. <doi:10.1007/s11222-009-9150-y>
Malsiner-Walli, G., Frühwirth-Schnatter, S. and Grün, B. (2016). Model-based clustering based on sparse finite Gaussian mixtures. Statistics and Computing 26, 303–324. <doi:10.1007/s11222-014-9500-2>
Sethuraman, A.J. (1994). A Constructive Definition of Dirichlet Priors, Statistica Sinica, 4, 639–650.
Examples
set.seed(123)
y <- c(rnorm(40,0,0.3), rnorm(20,5,0.3))
g <- c(rep(1,30), rep(2, 30))
plot(density(y[g==1]), xlim = c(-5,10))
lines(density(y[g==2]), col = 2)
out <- sample_fiSAN_sparsemix(nrep = 500, burn = 200, y = y, group = g,
nclus_start = 2,
maxK = 20, maxL = 20,
beta = 0.01)
out