wrMisc (version 1.2.3)

lmSelClu: Run lm on segmented data (from clustering)

Description

lmSelClu runs linear regression on data segmented previously (eg by clustering). This functio offers various types of (2-coefficient) linear regression on 2 columns of 'dat' (matrix with 3rd col named 'clu' or 'cluID', numeric elements for cluster-number). If argument 'clu' is (default) 'max', the column 'clu' will be inspected to take most frequent value of 'clu', otherwise a numeric entry specifying the cluster to extract is expected. Note: this function was initially made for use with results from diagCheck() Note: this function lacks means of judging godness of fit of the regression preformed & means for plotting

Usage

lmSelClu(
  dat,
  useCol = 1:2,
  clu = "max",
  regTy = "lin",
  filt1 = NULL,
  filt2 = NULL,
  silent = FALSE,
  callFrom = NULL
)

Arguments

dat

matrix or data.frame

useCol

(integer or charcter) specify which 2 columns of 'dat' to use for linear regression

clu

(character) name of cluster to be extracted and treatad

regTy

(character) change type used for linear regression : 'lin' for 1st col ~ 2nd col, 'res' for residue ~ 2nd col, 'norRes' for residue/2nd col ~2nd col or 'sqNorRes','inv' for 1st col ~ 1/(2nd col), 'invRes' for residue ~ 1/(2nd col)

filt1

(logical or numerical) filter criteria for 1st of 'useCol' , if numeric then select all lines of dat less than max of filt1

filt2

(logical or numerical) filter criteria for 2nd of 'useCol' , if numeric then select all lines of dat less than max of filt2

silent

(logical) suppress messages

callFrom

(character) allows easier tracking of message(s) produced

Value

lm object (or NULL if no data left)

See Also

lm

Examples

# NOT RUN {
set.seed(2016); ran1 <- runif(220)
mat1 <- round(rbind(matrix(c(1:100+ran1[1:100],rep(1,50)),ncol=3), 
  matrix(c(1:60,68:9+ran1[101:160],rep(2,60)),nc=3)),1)
colnames(mat1) <- c("a","BB","clu")
lmSelClu(mat1)
plot(mat1[which(mat1[,3]=="2"),1:2],col=grey(0.6))
abline(lmSelClu(mat1),lty=2,lwd=2)
# 
mat2 <- round(rbind(matrix(c(1:100+ran1[1:100],rep(1,50)),ncol=3), 
  matrix(c(1:60,(2:61+ran1[101:160])^2,rep(2,60)),nc=3)),1)
colnames(mat2) <- c("a","BB","clu")
(reg2 <- lmSelClu(mat2,regTy="sqNor"))
plot(function(x) coef(reg2)[2]+ (coef(reg2)[2]*x^2),xlim=c(1,70))
points(mat2[which(mat2[,3]=="2"),1:2],col=2)
# }