mccut: Evaluates a Two-Dimensional Monte Carlo Model in a Loop.

Description

evalmccut evaluates a Two-Dimensional Monte Carlo model using a loop on the uncertainty dimension. Within each loop, it calculates statistics in the variability dimension and stores them for further analysis. It allows to evaluate very high dimension models using (unlimited?) time instead of (limited) memory.

mcmodelcut builds a mcmodelcut object that can be sent to evalmccut.

Usage

evalmccut(model, nsv=ndvar(), nsu=ndunc(), seed=NULL, ind="index")
# S3 method for mccut
print(x, lim=c(0.025, 0.975), digits=3, ...)
mcmodelcut(x, is.expr=FALSE)

Value

An object of class mccut. This is a list including statistics evaluated within the third block. Each list consists of all the nsu values obtained. The print.mccut method print the median, the mean, the lim quantiles estimated on each statistics on the uncertainty dimension.

Arguments

model: a mcmodelcut object obtained using mcmodelcut function or (directly) a valid call including three blocks. See Details and Examples for the structure of the call.
x: a call or an expression (if is.expr=TRUE) including three blocks. See Details and Examples for the structure of the call.
nsv: The number of simulations for variability used in the evaluation.
nsu: The number of simulations for uncertainty used in the evaluation.
seed: The random seed used for the evaluation. If NULL the seed is unchanged.
ind: The variable name used in model to refers to the uncertainty. see Details and Example.
is.expr: FALSE to send a call, TRUE to send an expression (see mcmodel examples)
lim: A vector of values used for the quantile function (uncertainty dimension).
digits: Number of digits in the print.
...: Additional arguments to be passed in the final print function.

Details

This function should be used for high dimension Two-Dimensional Monte-Carlo simulations, when the memory limits of R are attained. The use of a loop will take (lots of) time, but less memory.

x (or model if a call is used directly in evalmccut) should be built as three blocks, separated by {.

The first block is evaluated once (and only once) before the first loop (step 1).
The second block, which should lead to an mc object, is evaluated using nsu = 1 (step 2).
The third block is evaluated on the mc object. All resulting statistics are stored (step 3).
The steps 2 and 3 are repeated nsu times. At each iteration, the values of the loop index (from 1 to nsu) is given to the variable specified in ind.
Finally, the nsu statistics are returned in an invisible object of class mccut.

Understanding this, the call should be built like this: {{block 1}{block 2}{block 3}}

The first block (maybe empty) is an expression that will be evaluated only once. This block should evaluate all "V" mcnode and "0" mcnodes. It may evaluate and "U" mcnode that will be sent in the second and third block by column, and, optionaly, some other codes (even "VU" mcnode, sent by columns) that can not be evaluated if ndunc=1 (e.g. sampling without replacement in the uncertainty dimension).
The second block is an expression that leads to the mc object. It must end with an expression as mymc <- mc(...). The variable specified as ind may be helpful to refer to the uncertainty dimension in this step
The last block should build a list of statistics refering to the mc object. The function summary should be used if a summary, a tornado on uncertainty (tornadounc.mccut) or a convergence diagnostic converg is needed, the function plot.mc should be used if a plot is needed, the function tornado should be used if a tornado is needed. Moreover, any other function that leads to a vector, a matrix, or a list of vector/matrix of statistics evaluated from the mc object may be used. list are time consuming.

IMPORTANT WARNING: do not forget to affect the results, since the print method provide only a summary of the results while all data may be stored in an mccut object.

Examples

Run this code

modEC3 <- mcmodelcut({

## First block:
## Evaluates all the 0, V and U nodes.
 { cook <- mcstoc(rempiricalD, type = "V", values = c(0, 1/5, 
 1/50), prob = c(0.027, 0.373, 0.6))
 serving <- mcstoc(rgamma, type = "V", shape = 3.93, rate = 0.0806)
 conc <- mcstoc(rnorm, type = "U", mean = 10, sd = 2)
 r <- mcstoc(runif, type = "U", min = 5e-04, max = 0.0015)
 }
## Second block:
## Evaluates all the VU nodes
## Leads to the mc object. 
 {
 expo <- conc * cook * serving
 dose <- mcstoc(rpois, type = "VU", lambda = expo)
 risk <- 1 - (1 - r)^dose
 res <- mc(conc, cook, serving, expo, dose, r, risk)
 }
## Third block:
## Leads to a list of statistics: summary, plot, tornado
## or any function leading to a vector (et), a list (minmax), 
## a matrix or a data.frame (summary)
 {
 list(
 sum = summary(res), 
 plot = plot(res, draw=FALSE), 
 minmax = lapply(res, range)
 )
 }
})

x <- evalmccut(modEC3, nsv = 101, nsu = 101, seed = 666)
summary(x)