VGAM (version 1.0-2)

calibrate.qrrvglm.control: Control Function for CQO/CAO Calibration

Description

Algorithmic constants and parameters for running calibrate.qrrvglm are set using this function.

Usage

calibrate.qrrvglm.control(object, trace = FALSE, Method.optim = "BFGS", gridSize = if (Rank == 1) 9 else 5, varI.latvar = FALSE, ...)

Arguments

object

The fitted CQO/CAO model. The user should ignore this argument.

trace

Logical indicating if output should be produced for each iteration. It is a good idea to set this argument to be TRUE since the computations are expensive.

Method.optim

Character. Fed into the method argument of optim.

gridSize

Numeric, recycled to length Rank. Controls the resolution of the grid used for initial values. For each latent variable, an equally spaced grid of length gridSize is cast from the smallest site score to the largest site score. Then the likelihood function is evaluated on the grid, and the best fit is chosen as the initial value. Thus increasing the value of gridSize increases the chance of obtaining the global solution, however, the computing time increases proportionately.

varI.latvar

Logical. For CQO objects only, this argument is fed into Coef.qrrvglm.

...

Avoids an error message for extraneous arguments.

Value

A list which with the following components.

trace

Numeric (even though the input can be logical).

gridSize

Positive integer.

varI.latvar

Logical.

Details

Most CQO/CAO users will only need to make use of trace and gridSize. These arguments should be used inside their call to calibrate.qrrvglm, not this function directly.

References

Yee, T. W. (2013) On constrained and unconstrained quadratic ordination. Manuscript in preparation.

See Also

calibrate.qrrvglm, Coef.qrrvglm.

Examples

## Not run:  hspider[, 1:6] <- scale(hspider[, 1:6])  # Needed when I.tol = TRUE
# set.seed(123)
# p1 <- cqo(cbind(Alopacce, Alopcune, Pardlugu, Pardnigr, 
#                 Pardpull, Trocterr, Zoraspin) ~
#           WaterCon + BareSand + FallTwig +
#           CoveMoss + CoveHerb + ReflLux,
#           family = poissonff, data = hspider, I.tol = TRUE)
# sort(deviance(p1, history = TRUE))  # A history of all the iterations
# 
# siteNos <- 3:4  # Calibrate these sites
# cp1 <- calibrate(p1, new = data.frame(depvar(p1)[siteNos, ]), trace = TRUE)
# ## End(Not run)

## Not run: 
# # Graphically compare the actual site scores with their calibrated values
# persp(p1, main = "Site scores: solid=actual, dashed=calibrated",
#       label = TRUE, col = "blue", las = 1)
# abline(v = latvar(p1)[siteNos], lty = 1,
#        col = 1:length(siteNos))  # Actual site scores
# abline(v = cp1, lty = 2, col = 1:length(siteNos))  # Calibrated values
# ## End(Not run)