setSeeds: setSeeds

In webb767/ppsmac: What the Package Does (One Line, Title Case)

Description

Brings saved random streams back to life. Reads a portable parallel seeds object (or file) and sets the seed collection in the environment.

Usage

setSeeds(projSeeds, run, verbose = FALSE)

Arguments

projSeeds	Required. Either an object of class portableSeeds (created by seedCreator) or a text string giving the name of an R saved file of the appropriate format (created by the seedCreator function).
run	Integer indicating which element from the portable seed collection is to be selected
verbose	Optional. Print out the state of the current generator. Default = FALSE.

Details

The portable seeds object is created by the function seedCreator. It is a list of lists. The list includes one set of initializing states for each separate run of a simulation. Within each of these sets, there will be enough information to initialize one or more streams of random numbers. These of "initializing states" are the internal states of CMRG random generators (see L'Ecuyer, 1999; L'Ecuyer, et al, 2002).

This function scans the project's portable parallel seeds (either an in-memory object or a named file), selects the desired run, and then it writes 3 variables into the environment called ".pps". 1) startStates is a collection of random generator states, one for each random stream from which the user might wish to draw random numbers. This is a fixed value which should not be altered during the program. It can be used to reset the generators to their initial positions. 2) currentStates is the collection of random generator states that will be updated. When the program calls the useStream function, the currentStates vector is updated. 3) currentStream indicates which of the currentStates should be used to draw the next random value.

At the outset, startStates and currentStates are identical and currentStream equals 1 (meaning the first element of currentStates is taken as the state of the random generator).

Value

nothing is returned. This function is used for the side effect of setting three objects in the global environment, the startStates (list), currentStates (list), and currentStream (an integer).

Author(s)

Paul E. Johnson [e-mail]

See Also

seedCreator to generate the input file for this function and useStream to change from one stream to another.

Examples

library(portableParallelSeeds)

projSeeds <- seedCreator(2000, 3, seed = 123456, file = "fruits.rds")
A1 <- projSeeds[[787]]
A1 ## shows states of 3 generators for run 787


setSeeds(projSeeds, run = 787, verbose = TRUE)
.Random.seed
getCurrentStream()
runif(4)

## read from file, take run 787's seed
myFruitySeeds <- readRDS("fruits.rds")
B1 <- myFruitySeeds[[787]]

identical(A1, B1) # check


setSeeds("fruits.rds", run=787)
.Random.seed
runif(4)


runOneSimulation <- function(run, streamsource, N, m, sd){
    setSeeds(streamsource, run = run, verbose= FALSE)
    datX <- rnorm(N, mean = m, sd = sd)
    datXmean <- mean(datX)
    useStream(2)
    datY <- rpois(N, lambda = m)
    datYmean <- mean(datY)
    useStream(1)
    datXplusOne <- rnorm(1, mean = m, sd = sd)
    ## Should be N+1'th element from first stream
    c("datXmean" = datXmean, "datYmean" = datYmean, "datXplusOne" = datXplusOne)
}


## Give seed collection object to each simulation, let each pick desired seed
    serial1 <- lapply(1:1000, runOneSimulation, projSeeds, N=800, m = 14, sd = 10.1)
    
    
    ## First re-load the seed object, then give to simulations
    fruits2 <- readRDS("fruits.rds")
    serial2 <- lapply(1:1000, runOneSimulation, fruits2, N=800, m = 14, sd = 10.1)
    
    
    ## Re-load file separately in each run
    serial3 <- lapply(1:1000, runOneSimulation, "fruits.rds", N = 800, m = 14, sd = 10.1)
    identical(serial1, serial2)
    identical(serial1, serial3)
    ## The results form the 3 lapply statements are all identical.

    ## Lets check run 912, you'll see what I mean.
    serial1[[912]]
    serial2[[912]]
    
    
    ## Next question. Can we re-set the current interactive session's seed
    ## and re-draw that same value?

    ## Put the seeds from run 912 back into the current session
    setSeeds("fruits.rds", run = 912, verbose = FALSE)

    ## Draw 801 elements    
    X <- rnorm(801, m=14, sd = 10.1)
    ## The 801'th element should be same as saved in the third element
    ## of the serial1[[912]] object.
    X[801]
    
    if(!all.equal(unname(X[801] - serial1[[912]][3]), 0)){
        stop("The difference is not 0, something went wrong")
    } else {
        print("Celebrate. The value in X[801] equals the serial1[[912]][3] value")
    }
    
    ##Bingo. I'm right. Can draw a understandably replicatable streams of
    ## random numbers, whether we draw 800, switch to a different stream,
    ## and then change back to draw another, or if we just draw 801 in one
    ## series.
    
    unlink("fruits.rds") #delete file

webb767/ppsmac documentation built on July 28, 2020, 12:59 a.m.