R help - Sep 2004 - Fixes for pls.pcr 0.2.2 plot.mvr("prediction")

Hi,

I just sent the following bugfixes to author Ron Wehrens, but thought I 
should also make them available here.  This fixes the prediction plots, 
which are currently erroneous.

======================

I found some bugs in the the plot.mvr() function, in the 
plottype="prediction" section:


1. Needs to set nscreens <- 1 for condition where length(which) = 1 AND 
length(nlv) = 1.  Otherwise, for such condition, the function exits with 
message "Error: Object nscreens not found".  Here is your code
snippet, and
my suggested addition is indicated by
========>>>>>>>>>>:

        if (length(which) > 1) {
            if (length(nlv) > 3) {
                par(ask = TRUE)
                par(mfrow = c(3, 2))
                nScreens <- ceiling(length(nlv)/3)
            }
            else {
                nScreens <- 1
                par(mfrow = c(length(nlv), 2))
            }
        }
        else {
            if (length(nlv) > 1) {
                nrow <- floor(sqrt(length(nlv)))
                mfrow <- c(nrow, ceiling(length(nlv)/nrow))
                par(mfrow = mfrow)
                nScreens <- 1
            }
 ======>>>>>nScreens <- 1
        }


2.  The prediction plots are being drawn based on incorrect indexing of the 
Ypred matrix -- this can be quite misleading :-)  Here is your code 
snippet:  Basically, need to replace all occurrences of
   mvrmodel$training$Ypred[, i, j] or mvrmodel$validat$Ypred[, i, j]
WITH
   mvrmodel$training$Ypred[, i, nlv[j]] or mvrmodel$validat$Ypred[, i, 
nlv[j]]



                for (j in indices) {
                  if (show[1]) {
                    plot(mvrmodel$Y[, i], mvrmodel$training$Ypred[,
                      i, j], xlab = "Measured", ylab =
"Predicted",
                      type = "n", xlim = range(mvrmodel$Y[, i],
                        mvrmodel$training$Ypred[, i, j]), ylim = 
range(mvrmodel$Y[,
                        i], mvrmodel$training$Ypred[, i, j]),
                      main = paste("Training data\n", nlv[j],
                        "latent variables"))
                    if (npred > 1)
                      mtext(ynames[i])
                    abline(0, 1, col = "blue")
                    points(mvrmodel$Y[, i], mvrmodel$training$Ypred[,
                      i, j])
                  }
                  if (show[2]) {
                    plot(mvrmodel$Y[, i], mvrmodel$validat$Ypred[,
                      i, j], xlab = "Measured", ylab =
"Predicted",
                      type = "n", xlim = range(mvrmodel$Y[, i],
                        mvrmodel$validat$Ypred[, i, j]), ylim = 
range(mvrmodel$Y[,
                        i], mvrmodel$validat$Ypred[, i, j]),
                      main = paste("Cross-validation data\n",
                        nlv[j], "latent variables"))
                    if (npred > 1)
                      mtext(ynames[i])
                    abline(0, 1, col = "blue")
                    points(mvrmodel$Y[, i], mvrmodel$validat$Ypred[,
                      i, j])


Regards,





Frank J. Iannarilli, Jr.  frankeye at cox.net
www.aerodyne.com/cosr/cosr.html