paircormin: Minimize Pairwise Correlation

In simplexdesign: Simplex Design for Stochastic Simulations and Agent Based Models

Description

Optimizes run order within classes of homogeneous agents to minimize pairwise correlation using Simulated Annealing.

Usage

paircormin(Design, kvec, ti = 10, tf = 0.0001, alph = 0.99, iter = 2000)

Arguments

Design	The unoptimized design
kvec	A vector containing the number of columns in each block of the design
ti	Initial temperature value for SA
tf	Final temperature value for SA
alph	Decay parameter for SA (should be less than 1)
iter	Number of iterations at each temperature value, should increase with design size

Details

This function optimizes run order within each block of a design made of multiple simplexes to minimize pairwise correlations using a Simulated Annealing algorithm. The criterion of interest is the sum of squared cor(xi,xj) for all pairs of columns between classes. It is recommended that the parameters (such as ti, tf, and iter) are scaled with the size of the design size, and that the criterion values should settle to some local optimal value.

Value

Design

The design optimized to minimize pairwise correlations between input classes.

CritVals

The pairwise correlations at each temperature change, useful for checking if the SA algorithm has converged.

Examples

# Generate Unoptimized Design
D1 <- MmSimplex(3,30,10,cords = 1,randst = 1,phival = 50)
D2 <- MmSimplex(3,30,10,cords = 1,randst = 1,phival = 50)
D3 <- MmSimplex(3,30,10,cords = 1,randst = 1,phival = 50)
D <- cbind(D1[[1]],D2[[1]],D3[[1]])
c1 <- sum(cor(D1[[1]])[upper.tri(cor(D1[[1]]))]^2)
c2 <- sum(cor(D2[[1]])[upper.tri(cor(D2[[1]]))]^2)
c3 <- sum(cor(D3[[1]])[upper.tri(cor(D3[[1]]))]^2)
## Sum all pariwise correlations and
sum(cor(D)[upper.tri(cor(D))]^2) - (c1 + c2 + c3)
## Optimize run order
Dopt <- paircormin(D, c(3,3,3), ti = 0.01, tf = 0.0005, iter = 25)
sum(cor(Dopt[[1]])[upper.tri(cor(Dopt[[1]]))]^2) - (c1 + c2 + c3)

simplexdesign documentation built on March 16, 2020, 5:06 p.m.