Replace chondro with faux_cell in hyperSpec.Rmd: Ran into an error with qr.solve

[bryanhanson]

I had to revert my conversion of instances of chondro to faux_cell in hyperSpec.Rmd because this code:

find.max <- function(y, x) {
  pos <- which.max(y) + (-1:1)
  X <- x[pos] - x[pos[2]]
  Y <- y[pos] - y[pos[2]]

  X <- cbind(1, X, X^2)
  coef <- qr.solve(X, Y)

  - coef[2] / coef[3] / 2 + x[pos[2]]
}

bandpos <- apply(chondro[[, , 990 ~ 1020]], 1, find.max, wl(chondro[, , 990 ~ 1020]))
refpos <- find.max(colMeans(chondro[[, , 990 ~ 1020]]),  wl(chondro[, , 990 ~ 1020]))

shift1 <- refpos - bandpos

throws an error from qr.solve. I also changed the relevant wavelengths; here is a failing example:

library("hyperSpec")

find.max <- function(y, x) {
  pos <- which.max(y) + (-1:1)
  X <- x[pos] - x[pos[2]]
  Y <- y[pos] - y[pos[2]]

  X <- cbind(1, X, X^2)
  coef <- qr.solve(X, Y)

  - coef[2] / coef[3] / 2 + x[pos[2]]
}

bandpos <- apply(faux_cell[[, , 1150 ~ 1250]], 1, find.max, wl(faux_cell[, , 1150 ~ 1250]))

Looking around a bit, I think it may have to do with the amplitudes and/or noise in faux_cell but I'm not sure, and a little experimentation did not reveal any insights (we are fitting to only 3 data points). Someone with more experience with qr.solve may need to take a look. There's also a note above the code in the vignette to use vanderMonde; I think this refers to X <- cbind(1, X, X^2) but I don't see how using vanderMonde gives an advantage here, except that one could make a hyperSpec object out of it, but that seems like overkill.

[GegznaV]

Due to this issue, two chunks of code in hyperSpec.Rmd are not functional and are temporary not executed. In the vignette, FIXME tag was added to highlight the issue.