You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
95 lines
4.2 KiB
R
95 lines
4.2 KiB
R
source("/home/taha/chepec/chetex/common/R/common.R")
|
|
|
|
##################################################
|
|
################## Ramanpk #######################
|
|
##################################################
|
|
Ramanpk <- function(Raman.exp, kerpk = 1, fitmaxiter = 50, gam = 0.6, scl.factor = 0.1) {
|
|
|
|
Raman.base <- baselinefit(Raman.exp, tau = 2.0, gam = gam, scl.factor = scl.factor, maxwdth = 200)
|
|
|
|
# This loop deals with the output from baselinefit()
|
|
# It makes a "melted" dataframe in long form for each
|
|
# separated peak for some baseline parameters
|
|
Raman.pks <- data.frame()
|
|
Raman.pks.basl <- data.frame()
|
|
Raman.pks.pmg <- data.frame()
|
|
Raman.pks.spl <- data.frame()
|
|
peaks <- 1:length(Raman.base$npks)
|
|
for (s in peaks) {
|
|
# recorded data in long form by separated peak
|
|
Raman.pks <- rbind(Raman.pks, # column names assigned after loop
|
|
data.frame(peak = factor(peaks[s]),
|
|
kernel = NA,
|
|
Raman.exp[Raman.base$indlsep[s]:Raman.base$indrsep[s], ]))
|
|
# the calculated baseline in long form by separated peak
|
|
Raman.pks.basl <- rbind(Raman.pks.basl,
|
|
data.frame(peak = factor(peaks[s]),
|
|
kernel = NA,
|
|
x = Raman.exp[Raman.base$indlsep[s]:Raman.base$indrsep[s]],
|
|
y = Raman.base$baseline$basisl[Raman.base$indlsep[s]:Raman.base$indrsep[s]]))
|
|
# the taut string estimation in long form by separated peak
|
|
Raman.pks.pmg <- rbind(Raman.pks.pmg,
|
|
data.frame(peak = factor(peaks[s]),
|
|
kernel = NA,
|
|
x = Raman.exp[Raman.base$indlsep[s]:Raman.base$indrsep[s]],
|
|
y = Raman.base$pmg$fn[Raman.base$indlsep[s]:Raman.base$indrsep[s]]))
|
|
# the weighted smoothed spline in long form by separated peak
|
|
Raman.pks.spl <- rbind(Raman.pks.spl,
|
|
data.frame(peak = factor(peaks[s]),
|
|
kernel = NA,
|
|
x = Raman.exp[Raman.base$indlsep[s]:Raman.base$indrsep[s]],
|
|
y = Raman.base$spl$reg[Raman.base$indlsep[s]:Raman.base$indrsep[s]]))
|
|
}
|
|
# Column names assigned to d.pks
|
|
names(Raman.pks) <- c("peak", "kernel", "x", "y")
|
|
|
|
|
|
# This loop calls pkdecompint() on each peak separately
|
|
# It makes a "melted" dataframe in long form for:
|
|
Raman.fit <- list() # holds pkdecompint output
|
|
Raman.fit.fitpk <- data.frame() # contains fitting curves
|
|
Raman.fit.parpk <- data.frame() # physical parameters by peak and kernel
|
|
Raman.nobasl <- data.frame() # data with baseline removed
|
|
peaks <- 1:length(Raman.base$npks)
|
|
for (s in peaks) {
|
|
######## PKDECOMPINT ########
|
|
if (length(kerpk) > 1) {
|
|
# set number of kernels per peak manually
|
|
Raman.fit[[s]] <- pkdecompint(Raman.base, intnum = s,
|
|
k = kerpk[s], maxiter = fitmaxiter)
|
|
} else {
|
|
# use number of kernels determined by baselinefit()
|
|
Raman.fit[[s]] <- pkdecompint(Raman.base, intnum = s,
|
|
k = Raman.base$npks[s], maxiter = fitmaxiter)
|
|
}
|
|
# Setup the dataframe that makes up the peak table
|
|
for (kernel in 1:Raman.fit[[s]]$num.ker) {
|
|
Raman.fit.parpk <- rbind(Raman.fit.parpk,
|
|
data.frame(peak = factor(Raman.fit[[s]]$intnr),
|
|
kernel = factor(kernel),
|
|
x = Raman.fit[[s]]$parpks[kernel, "loc"],
|
|
height = Raman.fit[[s]]$parpks[kernel, "height"],
|
|
area = Raman.fit[[s]]$parpks[kernel, "intens"],
|
|
fwhm = Raman.fit[[s]]$parpks[kernel, "FWHM"],
|
|
m = Raman.fit[[s]]$parpks[kernel, "m"],
|
|
accept = Raman.fit[[s]]$accept))
|
|
Raman.fit.fitpk <- rbind(Raman.fit.fitpk,
|
|
data.frame(peak = factor(peaks[s]),
|
|
kernel = factor(kernel),
|
|
x = Raman.fit[[s]]$x,
|
|
y = Raman.fit[[s]]$fitpk[kernel, ]))
|
|
}
|
|
Raman.nobasl <- rbind(Raman.nobasl,
|
|
data.frame(peak = factor(peaks[s]),
|
|
x = Raman.fit[[s]]$x,
|
|
y = Raman.fit[[s]]$y))
|
|
}
|
|
|
|
|
|
|
|
return(list(Raman.base = Raman.base,
|
|
Raman.peaks = Raman.pks,
|
|
Raman.fit.parpk = Raman.fit.parpk,
|
|
Raman.fit.fitpk = Raman.fit.fitpk,
|
|
Raman.nobasl = Raman.nobasl))
|
|
} |