glmmstan: Generalized liner mixed model in Rstan using glmer-style...

In norimune/glmmstan: Generalized liner mixed model in Rstan

Description

Using standard formula notation from glmer (lme4), defines a Stan model (rstan) and optionally samples from the posterior. Can optionally compute WAIC. Supports model families: "gaussian", "binomial", "poisson", "negative binomial", "beta", "gamma", "lognormal", "beta-binomial", "ordered",and "zero-inflated poisson, negative binomial and gamma".

Usage

glmmstan(formula_str, data, family="gaussian",
    center=FALSE, slice = NULL, offset=NULL,
    codeonly=FALSE, dataonly=FALSE, modelonly=FALSE,
    cauchy=2.5, lkj_corr=2,
    stancode=NULL, standata=NULL, stanmodel=NULL, stanfile=NULL, stanfit=NULL,
    parallel=FALSE, cores=NULL,
    iter=2000, warmup = NULL, chains= 2, thin=1)

Arguments

formula_str	Model formula or list of formulas, using glmer notation for varying effects.
data	Data frame or list.
family	Model family name or list of names for outcome(s). Valid choices are: "gaussian", "binomial" (logit link), "poisson" (log link), "nbinomial" (log link),"gamma" (log link), "beta" (logitlinks), "lognormal" (log link),"ordered" (cumulative logit),"beta-binomial"(logit link),"zipoisson"(log link),"zinbinomial"(log link),"zigamma"(log link).
center	If TRUE, center the independent variables(set means=0).
offset	Input offset term in model(only "poisson","nbinomial").
codeonly	If TRUE, output only stan code.
dataonly	If TRUE, output only dataset for stan.
modelonly	If TRUE, output only compiled stanmodel.
cauchy	Scale parameters of cauchy distribution which is prior distribution of variance components.
parallel	If TRUE, run the stanmodel in parallel using doparallel package.
cores	Number of cores to use when executing the chains in parallel.
iter	rstan parameter: total number of samples, including warmup.
warmup	rstan parameter: number of adaptation samples.
chains	rstan parameter: number of chains.
thin	rstan parameter: A positive integer specifying the period for saving sample.

Author(s)

Hiroshi Shimizu

See Also

glmer,stan

Examples

data(baseball)

#"gaussian"
fit1 <- glmmstan(salary_log~1,data=baseball,family="gaussian")
output_result(fit1) #output glmm result
output_stan(fit1) #output summarized stan result (including rhat index)
print(fit1) #output stan result (same print() in rstan)

#"lognormal" with random effect
fit2 <- glmmstan(salary~HR+(1+HR|team),data=baseball,family="lognormal")
output_result(fit2)$WAIC #output only WAIC

#"negative binomial" with offset term
fit3 <- glmmstan(HR~1,data=baseball,family="nbinomial",offset="ATbats")
output_result(fit3)$beta #output only coefficients and scale parameters

#"ordered" with centering indipendent variables
fit4 <- glmmstan(Cluster~salary,data=baseball,family="ordered",center=TRUE)
output_result(fit4)
output_code(fit4) #confirm the stan code

#output only stan code, datase, and stan model
code1 <- glmmstan(HR~1+(1|player),data=baseball,family="poisson",codeonly=TRUE)
dataset1 <- glmmstan(HR~1+(1|player),data=baseball,family="poisson",dataonly=TRUE)
model1 <- glmmstan(HR~1+(1|player),data=baseball,family="poisson",modelonly=TRUE)

fit5 <- stan(model_code=code1, data=dataset1)
fit6 <- sampling(model1,data=dataset1)

norimune/glmmstan documentation built on May 23, 2019, 9:33 p.m.