slice: Calculate likelihood "slices"

Description

Computes cross-section(s) of a multi-dimensional likelihood surface

Usage

slice(x, dim=1, ...)
sliceOld(fitted, which = 1:p, maxsteps = 100,
                       alpha = 0.01, zmax = sqrt(qchisq(1 - alpha/2, p)),
                       del = zmax/5, trace = FALSE,
                       tol.newmin=0.001, ...)
slice1D(params,fun,nt=101,lower=-Inf,
                    upper=Inf,verbose=TRUE, tranges=NULL,
                    fun_args = NULL,
                    ...)
slice2D(params,fun,nt=31,lower=-Inf,
                    upper=Inf,
                    cutoff=10,verbose=TRUE,
                    tranges=NULL,
                    ...)
slicetrans(params, params2, fun, extend=0.1, nt=401,
                       lower=-Inf, upper=Inf)

Value

An object of class slice with

slices

a list of individual parameter (or parameter-pair) slices, each of which is a data frame with elements

var1: name of the first variable
var2: (for 2D slices) name of the second variable
x: parameter values
y: (for 2D slices) parameter values
z: slice values
ranges: a list (?) of the ranges for each parameter
params: vector of baseline parameter values
dim: 1 or 2

sliceOld returns instead a list with elements profile and summary (see profile.mle2)

Arguments

x: a fitted model object of some sort
dim: dimensionality of slices (1 or 2)
params: a named vector of baseline parameter values
params2: a vector of parameter values
fun: an objective function
fun_args: additional arguments to pass to fun
nt: (integer) number of slice-steps to take
lower: lower bound(s) (stub?)
upper: upper bound(s) (stub?)
cutoff: maximum increase in objective function to allow when computing ranges
extend: (numeric) fraction by which to extend range beyond specified points
verbose: print verbose output?
fitted: A fitted maximum likelihood model of class “mle2”
which: a numeric or character vector describing which parameters to profile (default is to profile all parameters)
maxsteps: maximum number of steps to take looking for an upper value of the negative log-likelihood
alpha: maximum (two-sided) likelihood ratio test confidence level to find
zmax: maximum value of signed square root of deviance difference to find (default value corresponds to a 2-tailed chi-squared test at level alpha)
del: step size for profiling
trace: (logical) produce tracing output?
tol.newmin: tolerance for diagnosing a new minimum below the minimum deviance estimated in initial fit is found
tranges: a two-column matrix giving lower and upper bounds for each parameter
...: additional arguments (not used)

Author

Ben Bolker

Details

Slices provide a lighter-weight way to explore likelihood surfaces than profiles, since they vary a single parameter rather than optimizing over all but one or two parameters.

slice: is a generic method
slice1D: creates one-dimensional slices, by default of all parameters of a model
slice2D: creates two-dimensional slices, by default of all pairs of parameters in a model. In each panel the closed point represents the parameters given (typically the MLEs), while the open point represents the observed minimum value within the 2D slice. If everything has gone according to plan, these points should coincide (at least up to grid precision).
slicetrans: creates a slice along a transect between two specified points in parameter space (see calcslice in the emdbook package)

Examples

Run this code

x <- 0:10
y <- c(26, 17, 13, 12, 20, 5, 9, 8, 5, 4, 8)
d <- data.frame(x,y)
fit1 <- mle2(y~dpois(lambda=exp(lymax)/(1+x/exp(lhalf))),
   start=list(lymax=0,lhalf=0),
   data=d)
s1 <- bbmle::slice(fit1,verbose=FALSE)
s2 <- bbmle::slice(fit1,dim=2,verbose=FALSE)
require(lattice)
plot(s1)
plot(s2)
## 'transect' slice, from best-fit values to another point
st <- bbmle::slice(fit1,params2=c(5,0.5))
plot(st)

Run the code above in your browser using DataLab