Package 'clustTMB'

Title: Spatio-Temporal Finite Mixture Model using 'TMB'
Description: Fits a spatio-temporal finite mixture model using 'TMB'. Covariate, spatial and temporal random effects can be incorporated into the gating formula using multinomial logistic regression, the expert formula using a generalized linear mixed model framework, or both.
Authors: Andrea M. Havron [aut, cre, cph]
Maintainer: Andrea M. Havron <[email protected]>
License: GPL-3
Version: 0.1.0
Built: 2024-11-14 05:03:07 UTC
Source: https://github.com/andrea-havron/clusttmb

Help Index

Fit a finite mixture model using TMB

Description

Fit a finite mixture model using TMB

Usage

clustTMB(
  response = NULL,
  expertformula = ~1,
  gatingformula = ~1,
  expertdata = NULL,
  gatingdata = NULL,
  family = gaussian(link = "identity"),
  Offset = NULL,
  G = 2,
  rr = list(spatial = NULL, temporal = NULL, random = NULL),
  covariance.structure = NULL,
  Start = list(),
  Map = list(),
  initialization.args = list(control = init.options()),
  spatial.list = list(loc = NULL, mesh = NULL, init.range = list(gating.range = NULL,
    expert.range = NULL)),
  projection.dat = NULL,
  control = run.options()
)

Arguments

response

A numeric vector, matrix, or data frame of observations. When data are multivariate, rows correspond to observations and columns correspond to the multivariate response.
expertformula

Formula defining expert model. This formula corresponds to the covariates included in the response densities. Defaults to intercept only (~1) when no covariates are used.
gatingformula

Formula defining gating model. This formula corresponds to the covariates included in the mixing proportions (logistic regression). Defaults to intercept only (~1) when no covariates are used. When a random effects term is included in the gating network, this formula will be updated so that the intercept term is removed.
expertdata

Data frame containing expert model covariates.
gatingdata

Data frame containing gating model covariates.
family

Statistical distribution and link function of observations.
Offset

Constant in expertformula only used to offset density expectation.
G

Integer specifying the number of clusters.
rr

List specifying dimension of rank reduction in spatial, temporal, and/or random effects. Dimension must be smaller than the total dimension of the response. Rank reduction is applied only to the expertformula random effects. The rank reduction reduces the dimensionality of a correlated multivariate response to a smaller dimension independent response. When used, the covariance structure of the response is switched to 'Diagonal.' Defaults to NULL, no rank reduction. If rank reduction is used in conjunction with a random effect, that random effect must also be specified in the expert formula. Currently, rank reduction on temporal random effects is disabled.
covariance.structure

A character string specifying the covariance structure of the response using mclust naming scheme. See description of modelNames under ?Mclust for details.
Start

Set initial values for random effects parameters (fixed and random terms)
Map

Vector indicating parameter maps, see ?TMB::MakeADFun() for details. Defaults in clustTMB control this map argument and user input is limited
initialization.args

A list consisting of initialization settings used to generate initial values. control Calls init.options() to generate settings for initial values. Arguments of init.options() can be specified by the user.

  1. init.method - Single character string indicating initial clustering method. Methods include: hc, quantile, random, mclust, kmeans, mixed, user. Defaults to 'hc'. In the case where data are univariate and there are no covariates in the gating/expert formula, this defaults to 'quantile'

  2. hc.options - Named list of two character strings specifying hc modelName and hcUse when init.method = 'hc'. The default modelName is 'VVV' and the default use is 'SVD' unless gating/expert covariates specified, in which case default in VARS. See ?mclust::mclust.options for complete list of options.

  3. mix.method - String stating initialization method for mixed-type data (init.method = 'mixed'). Current default when Tweedie family specified. Options include: Gower kmeans (default), Gower hclust, and kproto.

  4. user - Numeric or character vector defining user specified initial classification. init.method must be set to 'user' when using this option.

spatial.list

List of data objects needed when fitting a spatial GMRF model
projection.dat

Spatial Points class of projection coordinates or Spatial Points Dataframe containing projection coordinates and projection covariates
control

List controlling whether models are run and whether standard errors are calculated.

Value

list of objects from fitted model

Examples

data("faithful")
m1 <- clustTMB(response = faithful, covariance.structure = "VVV")
plot(faithful$eruptions, faithful$waiting, pch = 16, col = m1$report$classification + 1)

Get fixed-effect coefficients

Description

Get fixed-effect coefficients

Usage

## S3 method for class 'clustTMB'
coef(object, complete = FALSE, ...)

Arguments

object

The fitted clustTMB model
complete

Currently ignored
...

Currently ignored

Value

names numeric vector

Extract the AIC of a clustTMB model

Description

Extract the AIC of a clustTMB model

Usage

## S3 method for class 'clustTMB'
extractAIC(fit, scale, k = 2, ...)

Arguments

fit

The fitted clustTMB model
scale

The scale, currently ignored
k

Penalization parameter, defaults to 2
...

Currently ignored

Value

numeric value

Fixed Covariance Structure names

Description

Fixed Covariance Structure names

Usage

fixStruct.names()

Value

character vector naming available fixed Covariance Structures

Examples

fixStruct.names()

Initialization options with S3 classes

Description

Initialization options with S3 classes

Usage

init.options(
  init.method = "hc",
  hc.options = list(modelName = "VVV", use = "SVD"),
  exp.init = list(mahala = TRUE),
  mix.method = "Gower kmeans",
  user.class = integer()
)

Arguments

init.method

Name of method used to set initial values. If init.method = 'user', must define 'user.class' with a classification vector.
hc.options

Model names and use when init.method is 'hc' following conventions of mclust::mclust.options()
exp.init

Turn on mahala initialization when expert network
mix.method

Initialization methods when data are mixed. Default method when data are Tweedie distributed.
user.class

Vector of classification vector set by user and required when init.method = 'user'

Value

list of initialization specifications

Examples

init.options()
init.options(init.method = "hc")
init.options(init.method = "mixed")
init.options(init.method = "user", user.class = c(1, 1, 2, 1, 3, 3, 1, 2))

Extract the log likelihood of a clustTMB model

Description

Extract the log likelihood of a clustTMB model

Usage

## S3 method for class 'clustTMB'
logLik(object, ...)

Arguments

object

The fitted clustTMB model
...

Currently ignored

Value

object of class logLik with attributes

Lognormal family and link specification

Description

Lognormal family and link specification

Usage

lognormal(link = "identity")

Arguments

link

link function association with family

Value

An object of class "family"

Examples

fam <- lognormal()
fam$family
fam$link

Apply classification method dependent on init.method

Description

Apply classification method dependent on init.method

Usage

mkInitClass(n.g, n.i, n.j, control, y)

Arguments

n.g

Number of clusters
n.i

Number of observations
n.j

Number of columns
control

Classification settings from init.options()
y

Observations

Value

classification vector

Examples

data("faithful")
mkInitClass(2, nrow(faithful), ncol(faithful), init.options(), faithful)

Parameter Information

Description

Parameter Information

Usage

parm.lookup()

Value

Description of parameters, including dimension and structure

Examples

parm.lookup()

Print brief model summary

Description

Invoke TMB's print.report function

Usage

## S3 method for class 'clustTMB'
print(x, ...)

Arguments

x

The fitted clustTMB model
...

Not used

Value

numeric matrix of parameter estimate and standard errors

Run Options

Description

Run Options

Usage

run.options(check.input = NULL, run.model = NULL, do.sdreport = NULL)

Arguments

check.input

TRUE: Return initial values before running TMB
run.model

FALSE: Return TMB object before optimizing model
do.sdreport

TRUE: Run delta method to obtain standard errors

Value

list

Examples

run.options()

Calculates skewness

Description

Calculates skewness

Usage

skewness(x)

Arguments

x

numeric vector of values for which skewness is calculated

Value

skewness value of x

Examples

skewness(rgamma(100, 1, 1))

Split formula containing special random effect terms

Description

Parse a formula into fixed formula and random effect terms, treating 'special' terms appropriately

Usage

splitForm(
  formula,
  defaultTerm = "norm",
  allowFixedOnly = TRUE,
  allowNoSpecials = TRUE,
  debug = FALSE
)

Arguments

formula

a formula containing special random effect terms
defaultTerm

default type for non-special RE terms
allowFixedOnly

(logical) are formulas with no RE terms OK?
allowNoSpecials

(logical) are formulas with only standard RE terms OK?
debug

(logical) debug?

Details

Taken from Steve Walker's lme4ord, ultimately from the flexLambda branch of lme4 https://github.com/stevencarlislewalker/lme4ord/blob/master/R/formulaParsing.R. Mostly for internal use.

Value

a list containing elements fixedFormula; reTrmFormulas list of x | g formulas for each term; reTrmAddArgs list of function+additional arguments, i.e. list() (non-special), foo() (no additional arguments), foo(addArgs) (additional arguments); reTrmClasses (vector of special functions/classes, as character)

Author(s)

Steve Walker

Examples

splitForm(~x+y)                     ## no specials or RE
splitForm(~x+y+(f|g))               ## no specials
splitForm(~x+y+diag(f|g))           ## one special
splitForm(~x+y+(diag(f|g)))         ## 'hidden' special
splitForm(~x+y+(f|g)+cs(1|g))       ## combination
splitForm(~x+y+(1|f/g))             ## 'slash'; term
splitForm(~x+y+(1|f/g/h))             ## 'slash'; term
splitForm(~x+y+(1|(f/g)/h))             ## 'slash'; term
splitForm(~x+y+(f|g)+cs(1|g)+cs(a|b,stuff))  ## complex special
splitForm(~(((x+y))))               ## lots of parentheses
splitForm(~1+rr(f|g,n=2))

summary tables of model parameters

Description

Invoke TMB's summary.sdreport function

Usage

## S3 method for class 'clustTMB'
summary(
  object,
  select = c("all", "fixed", "random", "report"),
  p.value = FALSE,
  ...
)

Arguments

object

The fitted clustTMB model
select

Parameter classes to select. Can be any subset of "fixed" (θ^\hat{\theta}), "random" (u^\hat{u}) or "report" (ϕ(u^,θ)^\phi(\hat{u},\hat{\theta)}) using notation as TMB::sdreport().
p.value

Add column with approximate p-values
...

Currently ignored

Value

numeric matrix of parameter estimate and standard errors

Tweedie family and link specification

Description

Tweedie family and link specification

Usage

tweedie(link = "log")

Arguments

link

link function association with family

Value

An object of class "family"

Examples

fam <- tweedie()
fam$family
fam$link