R/fittingloop.R
In danielcanueto/Dolphin: Reliable automatic profiling of 1D 1H NMR Spectra, with additional tools for analysis
#Workflow of optimization of metabolite and baseline signals parameters. There are several optimization iterations ("iter") and the one with less error is chosen. The maximum number of iterations depends on the complexity of the ROI. Then the half bandwidth and j coupling are optimized separately as they have diferent behavior with the least squares algorithm. After the first fitting there can be further ones, depending on the need to add additional signals to adapt the signals and ROI information provided by the user to the concrete characteristics of the spectrum.
fittingloop = function(FeaturesMatrix,Xdata,Ydata,program_parameters) {
#Preallocation of output and setting of necessary variables for loop
signals_parameters=rep(0,length(as.vector(t(FeaturesMatrix[, seq(1, 9, 2), drop = F]))))
iterrep = 0
fitting_maxiterrep = program_parameters$fitting_maxiterrep
signals_to_fit = which(FeaturesMatrix[, 11] != 0)
paramprov=rep(0,nrow(FeaturesMatrix)*5)
#Necessary information to incorporate additional signals if necessary
range_ind = round(program_parameters$additional_signal_ppm_distance / program_parameters$buck_step)
#Function where to find a minimum
residFun <-
function(par, observed, xx,multiplicities,roof_effect,freq,bins)
observed[bins] - colSums(signal_fitting(par, xx,multiplicities,roof_effect,freq))[bins]
# Loop to control if additional signals are incorporated, until a maximum of iterations specified bt fitting_maxiterrep.
# If at the last fitting the improvement was lesser than 25% respective to the previous fitting,
# iterrep becomes equal to fitting_maxiterrep and the loop is stooped
while (iterrep <= fitting_maxiterrep) {
iter = 0
errorprov = error1 = 3000
worsterror = 0
dummy = error2 = 3000
multiplicities=FeaturesMatrix[,11]
roof_effect=FeaturesMatrix[,12]
fitted_signals = signal_fitting(as.vector(t(FeaturesMatrix[,c(2,3,6,7,9)])),
Xdata,multiplicities,roof_effect,program_parameters$freq)
bins=c()
for (i in signals_to_fit) {
sorted_bins=sort(fitted_signals[i,]/sum(fitted_signals[i,]),decreasing=T,index.return=T)
if(length(sorted_bins$x)>0) {
bins2= sorted_bins$ix[1:which.min(abs(cumsum(sorted_bins$x)-0.9))]
distance=diff(FeaturesMatrix[i,3:4])/program_parameters$buck_step
distance2=round(diff(FeaturesMatrix[i,9:10])/program_parameters$buck_step/program_parameters$freq)
bins2=unique(as.vector(sapply(seq(distance),function(x)bins2-x)))
bins2=unique(bins2,min(bins2)-distance2,max(bins2)+distance2)
bins2=bins2[bins2>0&bins2<length(Xdata)]
# aa=peakdet(fitted_signals[i,],0.00001)$maxtab$pos
# aa=as.vector(sapply(seq(distance),function(x)aa-x))
# #
# lol=sapply(seq(distance),function(x)min(Ydata[bins2]-fitted_signals[i,(bins2-x)]))
# FeaturesMatrix[i,2]=FeaturesMatrix[i,2]+lol
bins=unique(c(bins,bins2))
}}
#Depending on the complexity of the ROI, more or less iterations are performed
if (is.numeric(program_parameters$fitting_maxiter)) {
fitting_maxiter = program_parameters$fitting_maxiter
} else {
if (nrow(FeaturesMatrix)> 8 |
any(FeaturesMatrix[, 4] - FeaturesMatrix[, 3] > 0.01)) {
fitting_maxiter = 20
} else if ((nrow(FeaturesMatrix)> 5 &&
nrow(FeaturesMatrix)< 9)) {
fitting_maxiter = 14
} else {
fitting_maxiter = 8
}
}
#Conditions to keep the loop:
# -The error is bigger than the specified in program_parameters
# -There is no fitting with more than 66.7% improvement from the worst solution
# -The loop has not arrived the specified maximum of iterations
while (error1 > program_parameters$errorprov &error1 > (1 / 3 * worsterror) & iter < fitting_maxiter) {
#Initialization of parameters to optimize. In every iteration the initialization will be different
lb = as.vector(t(FeaturesMatrix[, seq(1, 9, 2), drop = F]))
ub = as.vector(t(FeaturesMatrix[, seq(2, 10, 2), drop = F]))
set.seed(iter);s0 = lb + (ub - lb) * runif(length(ub))
order1=order(rowMeans(FeaturesMatrix[signals_to_fit ,3:4,drop=F])[signals_to_fit])
# aaa=iter%%3/3
# bbb=ifelse((iter+1)%%3/3==0,1,(iter+1)%%3/3)
# s0[which(seq_along(s0)%%5==2)]=lb[which(seq_along(s0)%%5==2)] + (ub[which(seq_along(s0)%%5==2)] - lb[which(seq_along(s0)%%5==2)]) * runif(1,min=aaa,max=bbb)
#During the first two iterations, find the peaks on the region of the spectrum. If the number of peaks is the same that the expected on the ROI and the location is similar, the signals are located where there are the peaks with minimum $chemical_shift tolerance.
peaks_xdata = peakdet(c(Ydata[1],diff(Ydata)), program_parameters$peakdet_minimum*0.1*max(1e-10,max(Ydata)),Xdata)
if (iter<4&length(peaks_xdata$maxtab$val)>0) {
peaks_bindata = peakdet(c(Ydata[1],diff(Ydata)), program_parameters$peakdet_minimum*0.1*max(1e-10,max(Ydata)))
peaks=peaks_xdata$maxtab$pos[sort(peaks_xdata$maxtab$val,decreasing=T,index.return=T)$ix[1:sum(multiplicities[signals_to_fit])]]
peaks_compare=rowMeans(FeaturesMatrix[signals_to_fit ,3:4,drop=F])
for (i in 1:length(peaks_compare)) {
ind=sort(abs(peaks-peaks_compare[i]),index.return=T)$ix[1:multiplicities[i]]
if (!is.na(mean(peaks[ind]))&&mean(peaks[ind])>FeaturesMatrix[i,3]&&mean(peaks[ind])<FeaturesMatrix[i,4]) {
s0[which(seq_along(s0)%%5==2)[i]]=mean(peaks[ind])
lb[which(seq_along(s0)%%5==2)[i]]=mean(peaks[ind])-0.001
ub[which(seq_along(s0)%%5==2)[i]]=mean(peaks[ind])+0.001
}
}
#Main optimization
set.seed(iter);nls.out <-
minpack.lm::nls.lm(
par = s0,
fn = residFun,
observed = Ydata,
xx = Xdata,
multiplicities=multiplicities,
roof_effect=roof_effect,
freq=program_parameters$freq,
lower = lb,
upper = ub,
control = minpack.lm::nls.lm.control(
factor = program_parameters$factor,
maxiter = program_parameters$nls_lm_maxiter,
ftol = program_parameters$ftol,
ptol = program_parameters$ptol
)
)
# #Procedure to calculate the fititng error in all the ROI
#An adapted MSE error is calculated, and the parameters of the optimization with less MSE are stored
iter = iter + 1
order2=order(coef(nls.out)[which(seq_along(coef(nls.out))%%5==2)][signals_to_fit])
errorprov = (sqrt(nls.out$deviance / length(Ydata))) * 100 / (max(Ydata) -min(Ydata))
if (is.nan(errorprov) || is.na(errorprov)) errorprov = error1
if (errorprov < error1 && identical(order1,order2)) {
error1 = errorprov
paramprov=coef(nls.out)
} else if (errorprov > worsterror) {
worsterror = errorprov
}
} else {
#If in the first two iterations the procedure of finding peaks is not effective enough, the irignal chemical $chemical_shift and chemical $chemical_shift tolerance of every signal is maintained
set.seed(iter);nls.out <-
minpack.lm::nls.lm(
par = s0,
fn = residFun,
observed = Ydata,
xx = Xdata,
multiplicities=multiplicities,
roof_effect=roof_effect,
freq=program_parameters$freq,
lower = lb,
upper = ub,
control = minpack.lm::nls.lm.control(
factor = program_parameters$factor,
maxiter = program_parameters$nls_lm_maxiter,
ftol = program_parameters$ftol,
ptol = program_parameters$ptol
)
)
iter = iter + 1
order2=order(coef(nls.out)[which(seq_along(coef(nls.out))%%5==2)][signals_to_fit])
# #Procedure to calculate the fititng error in all the ROI
#An adapted MSE error is calculated, and the parameters of the optimization with less MSE are stored
errorprov = (sqrt(nls.out$deviance / length(Ydata))) * 100 / (max(Ydata) -min(Ydata))
if (is.nan(errorprov) || is.na(errorprov))errorprov = error1
if (errorprov < error1 && identical(order1,order2)) {
error1 = errorprov
paramprov=coef(nls.out)
} else if (errorprov > worsterror) {
worsterror = errorprov
}
}}
signals_parameters = paramprov
fitted_signals = signal_fitting(signals_parameters,
Xdata,multiplicities,roof_effect,program_parameters$freq)
bins=c()
for (ind in signals_to_fit) {
sorted_bins=sort(fitted_signals[ind,]/sum(fitted_signals[ind, ]),decreasing=T,index.return=T)
if(length(sorted_bins$x)>0) bins= sorted_bins$ix[1:which.min(abs(cumsum(sorted_bins$x)-0.75))]
}
if (length(bins)==0) bins=seq_along(Ydata)
#Correction of half_bandwidth and j-coupling
iter = 0
error22=error2=error1
errorprov = error1=3000
#Only half_bandwidth and j-coupling will have different lower und upper bounds.
change_indexes=which(seq_along(lb)%%5!=3 & seq_along(lb)%%5!=4 & seq_along(lb)%%5!=0)
lb[change_indexes]=ub[change_indexes]=signals_parameters[change_indexes]
while (iter < 3) {
set.seed(iter);s0 = lb + (ub - lb) * runif(length(ub))
nls.out <-
minpack.lm::nls.lm(
par = s0,
fn = residFun,
observed = Ydata,
xx = Xdata,
multiplicities=multiplicities,
roof_effect=roof_effect,
freq=program_parameters$freq,
lower = lb,
upper = ub,
control = minpack.lm::nls.lm.control(
factor = program_parameters$factor,
maxiter = program_parameters$nls_lm_maxiter,
ftol = program_parameters$ftol,
ptol = program_parameters$ptol
)
)
iter = iter + 1
# #Procedure to calculate the fititng error in all the ROI
#An adapted MSE error is calculated, and the parameters of the optimization with less MSE are stored
errorprov = (sqrt(nls.out$deviance / length(Ydata))) * 100 / (max(Ydata) -
min(Ydata))
if (is.nan(errorprov) || is.na(errorprov)) errorprov = error1
if (errorprov < error1) {
error1 = errorprov
paramprov=coef(nls.out)
} else if (errorprov > worsterror) {
worsterror = errorprov
}
if (error1 < error2) {
error2 = error1
signals_parameters = paramprov
}
}
#If half_bandwidth and j-coup change improves fitting
iterrep = iterrep + 1
#If the fitting seems to be still clearly improvable through the addition of signals
if (iterrep <= fitting_maxiterrep& error22 < (program_parameters$additional_signal_improvement * dummy) &
(error22 > program_parameters$additional_signal_percentage_limit)&length(peaks_xdata$maxtab$pos)>sum(multiplicities[signals_to_fit])) {
# print('Trying to improve initial fit adding peaks')
#I find peaks on the residuals
residual_peaks = tryCatch(peakdet(c(nls.out$fvec[1],diff(nls.out$fvec)), program_parameters$peakdet_minimum*max(1e-10,max(Ydata))),error= function(e){
dummy=list(signals_parameters=signals_parameters,error1=error1)
return(dummy)
})
if (is.null(residual_peaks$maxtab) == F) {
#Preparation of information of where signals of interest are located
dummy=multiplicities[signals_to_fit]%%2
dummy[dummy==0]=2
additional_signal_matrix = matrix(paramprov,nrow(FeaturesMatrix),5,byrow = TRUE)
points_to_avoid = abs(rbind(matrix(Xdata,length(signals_to_fit),length(Xdata),byrow = TRUE) - matrix(
additional_signal_matrix[signals_to_fit, 2] - (additional_signal_matrix[signals_to_fit, 5]/dummy)/program_parameters$freq,length(signals_to_fit),length(Xdata)),
matrix(Xdata,length(signals_to_fit),length(Xdata),byrow = TRUE) - matrix(additional_signal_matrix[signals_to_fit, 2] + (additional_signal_matrix[signals_to_fit, 5]/dummy)/program_parameters$freq,length(signals_to_fit),length(Xdata))))
points_to_avoid = apply(points_to_avoid, 1, which.min)
seq_range = c()
for (i in (-range_ind):range_ind) seq_range = append(seq_range, points_to_avoid - i)
#Finding of posible additional signals to incorporate if there are not in zones where the signals o interest are located
residual_peaks=cbind(residual_peaks$maxtab$pos, residual_peaks$maxtab$val)[residual_peaks$maxtab$pos %in% peaks_bindata$maxtab$pos,,drop=F]
valid_residual_peaks = matrix(NA, 0, 2)
if (nrow(residual_peaks)>0) {
for (i in seq(nrow(residual_peaks))) {
if (any(abs(points_to_avoid - residual_peaks[i, 1]) < range_ind) == F) valid_residual_peaks = rbind(valid_residual_peaks, residual_peaks[i, ])
}
#Selection of more intense additional signals
if (nrow(valid_residual_peaks) > program_parameters$signals_to_add) {
ad = sort(valid_residual_peaks[, 2],decreasing = T,index.return = T)$ix
valid_residual_peaks = valid_residual_peaks[ad[1:min(program_parameters$signals_to_add, length(ad))], , drop = F]
}
#Creation of rows to incorporate to FeaturesMatrix
if (nrow(valid_residual_peaks)>0) {
dummy = t(replicate(nrow(valid_residual_peaks),FeaturesMatrix[1,]))
dummy[, 2] = Ydata[valid_residual_peaks[, 1]]
dummy[, 3] = Xdata[valid_residual_peaks[, 1]] - 0.001
dummy[, 4] = Xdata[valid_residual_peaks[, 1]] + 0.001
dummy[, 5]=min(FeaturesMatrix[,5])
dummy[, 6]=min(FeaturesMatrix[,6])
dummy[, c(9,10,12)] = rep(0, nrow(valid_residual_peaks))
dummy[, 11] = rep(1, nrow(valid_residual_peaks))
FeaturesMatrix = rbind(FeaturesMatrix, dummy)
} else {
iterrep = fitting_maxiterrep +1
}
} else {
iterrep = fitting_maxiterrep +1
}}
} else {
iterrep = fitting_maxiterrep +1
}
}
optim_parameters=list(signals_parameters=signals_parameters,error1=error1)
return(optim_parameters)
}
danielcanueto/Dolphin documentation built on May 14, 2019, 3:51 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
#Workflow of optimization of metabolite and baseline signals parameters. There are several optimization iterations ("iter") and the one with less error is chosen. The maximum number of iterations depends on the complexity of the ROI. Then the half bandwidth and j coupling are optimized separately as they have diferent behavior with the least squares algorithm. After the first fitting there can be further ones, depending on the need to add additional signals to adapt the signals and ROI information provided by the user to the concrete characteristics of the spectrum.
fittingloop = function(FeaturesMatrix,Xdata,Ydata,program_parameters) {
#Preallocation of output and setting of necessary variables for loop
signals_parameters=rep(0,length(as.vector(t(FeaturesMatrix[, seq(1, 9, 2), drop = F]))))
iterrep = 0
fitting_maxiterrep = program_parameters$fitting_maxiterrep
signals_to_fit = which(FeaturesMatrix[, 11] != 0)
paramprov=rep(0,nrow(FeaturesMatrix)*5)
#Necessary information to incorporate additional signals if necessary
range_ind = round(program_parameters$additional_signal_ppm_distance / program_parameters$buck_step)
#Function where to find a minimum
residFun <-
function(par, observed, xx,multiplicities,roof_effect,freq,bins)
observed[bins] - colSums(signal_fitting(par, xx,multiplicities,roof_effect,freq))[bins]
# Loop to control if additional signals are incorporated, until a maximum of iterations specified bt fitting_maxiterrep.
# If at the last fitting the improvement was lesser than 25% respective to the previous fitting,
# iterrep becomes equal to fitting_maxiterrep and the loop is stooped
while (iterrep <= fitting_maxiterrep) {
iter = 0
errorprov = error1 = 3000
worsterror = 0
dummy = error2 = 3000
multiplicities=FeaturesMatrix[,11]
roof_effect=FeaturesMatrix[,12]
fitted_signals = signal_fitting(as.vector(t(FeaturesMatrix[,c(2,3,6,7,9)])),
Xdata,multiplicities,roof_effect,program_parameters$freq)
bins=c()
for (i in signals_to_fit) {
sorted_bins=sort(fitted_signals[i,]/sum(fitted_signals[i,]),decreasing=T,index.return=T)
if(length(sorted_bins$x)>0) {
bins2= sorted_bins$ix[1:which.min(abs(cumsum(sorted_bins$x)-0.9))]
distance=diff(FeaturesMatrix[i,3:4])/program_parameters$buck_step
distance2=round(diff(FeaturesMatrix[i,9:10])/program_parameters$buck_step/program_parameters$freq)
bins2=unique(as.vector(sapply(seq(distance),function(x)bins2-x)))
bins2=unique(bins2,min(bins2)-distance2,max(bins2)+distance2)
bins2=bins2[bins2>0&bins2<length(Xdata)]
# aa=peakdet(fitted_signals[i,],0.00001)$maxtab$pos
# aa=as.vector(sapply(seq(distance),function(x)aa-x))
# #
# lol=sapply(seq(distance),function(x)min(Ydata[bins2]-fitted_signals[i,(bins2-x)]))
# FeaturesMatrix[i,2]=FeaturesMatrix[i,2]+lol
bins=unique(c(bins,bins2))
}}
#Depending on the complexity of the ROI, more or less iterations are performed
if (is.numeric(program_parameters$fitting_maxiter)) {
fitting_maxiter = program_parameters$fitting_maxiter
} else {
if (nrow(FeaturesMatrix)> 8 |
any(FeaturesMatrix[, 4] - FeaturesMatrix[, 3] > 0.01)) {
fitting_maxiter = 20
} else if ((nrow(FeaturesMatrix)> 5 &&
nrow(FeaturesMatrix)< 9)) {
fitting_maxiter = 14
} else {
fitting_maxiter = 8
}
}
#Conditions to keep the loop:
# -The error is bigger than the specified in program_parameters
# -There is no fitting with more than 66.7% improvement from the worst solution
# -The loop has not arrived the specified maximum of iterations
while (error1 > program_parameters$errorprov &error1 > (1 / 3 * worsterror) & iter < fitting_maxiter) {
#Initialization of parameters to optimize. In every iteration the initialization will be different
lb = as.vector(t(FeaturesMatrix[, seq(1, 9, 2), drop = F]))
ub = as.vector(t(FeaturesMatrix[, seq(2, 10, 2), drop = F]))
set.seed(iter);s0 = lb + (ub - lb) * runif(length(ub))
order1=order(rowMeans(FeaturesMatrix[signals_to_fit ,3:4,drop=F])[signals_to_fit])
# aaa=iter%%3/3
# bbb=ifelse((iter+1)%%3/3==0,1,(iter+1)%%3/3)
# s0[which(seq_along(s0)%%5==2)]=lb[which(seq_along(s0)%%5==2)] + (ub[which(seq_along(s0)%%5==2)] - lb[which(seq_along(s0)%%5==2)]) * runif(1,min=aaa,max=bbb)
#During the first two iterations, find the peaks on the region of the spectrum. If the number of peaks is the same that the expected on the ROI and the location is similar, the signals are located where there are the peaks with minimum $chemical_shift tolerance.
peaks_xdata = peakdet(c(Ydata[1],diff(Ydata)), program_parameters$peakdet_minimum*0.1*max(1e-10,max(Ydata)),Xdata)
if (iter<4&length(peaks_xdata$maxtab$val)>0) {
peaks_bindata = peakdet(c(Ydata[1],diff(Ydata)), program_parameters$peakdet_minimum*0.1*max(1e-10,max(Ydata)))
peaks=peaks_xdata$maxtab$pos[sort(peaks_xdata$maxtab$val,decreasing=T,index.return=T)$ix[1:sum(multiplicities[signals_to_fit])]]
peaks_compare=rowMeans(FeaturesMatrix[signals_to_fit ,3:4,drop=F])
for (i in 1:length(peaks_compare)) {
ind=sort(abs(peaks-peaks_compare[i]),index.return=T)$ix[1:multiplicities[i]]
if (!is.na(mean(peaks[ind]))&&mean(peaks[ind])>FeaturesMatrix[i,3]&&mean(peaks[ind])<FeaturesMatrix[i,4]) {
s0[which(seq_along(s0)%%5==2)[i]]=mean(peaks[ind])
lb[which(seq_along(s0)%%5==2)[i]]=mean(peaks[ind])-0.001
ub[which(seq_along(s0)%%5==2)[i]]=mean(peaks[ind])+0.001
}
}
#Main optimization
set.seed(iter);nls.out <-
minpack.lm::nls.lm(
par = s0,
fn = residFun,
observed = Ydata,
xx = Xdata,
multiplicities=multiplicities,
roof_effect=roof_effect,
freq=program_parameters$freq,
lower = lb,
upper = ub,
control = minpack.lm::nls.lm.control(
factor = program_parameters$factor,
maxiter = program_parameters$nls_lm_maxiter,
ftol = program_parameters$ftol,
ptol = program_parameters$ptol
)
)
# #Procedure to calculate the fititng error in all the ROI
#An adapted MSE error is calculated, and the parameters of the optimization with less MSE are stored
iter = iter + 1
order2=order(coef(nls.out)[which(seq_along(coef(nls.out))%%5==2)][signals_to_fit])
errorprov = (sqrt(nls.out$deviance / length(Ydata))) * 100 / (max(Ydata) -min(Ydata))
if (is.nan(errorprov) || is.na(errorprov)) errorprov = error1
if (errorprov < error1 && identical(order1,order2)) {
error1 = errorprov
paramprov=coef(nls.out)
} else if (errorprov > worsterror) {
worsterror = errorprov
}
} else {
#If in the first two iterations the procedure of finding peaks is not effective enough, the irignal chemical $chemical_shift and chemical $chemical_shift tolerance of every signal is maintained
set.seed(iter);nls.out <-
minpack.lm::nls.lm(
par = s0,
fn = residFun,
observed = Ydata,
xx = Xdata,
multiplicities=multiplicities,
roof_effect=roof_effect,
freq=program_parameters$freq,
lower = lb,
upper = ub,
control = minpack.lm::nls.lm.control(
factor = program_parameters$factor,
maxiter = program_parameters$nls_lm_maxiter,
ftol = program_parameters$ftol,
ptol = program_parameters$ptol
)
)
iter = iter + 1
order2=order(coef(nls.out)[which(seq_along(coef(nls.out))%%5==2)][signals_to_fit])
# #Procedure to calculate the fititng error in all the ROI
#An adapted MSE error is calculated, and the parameters of the optimization with less MSE are stored
errorprov = (sqrt(nls.out$deviance / length(Ydata))) * 100 / (max(Ydata) -min(Ydata))
if (is.nan(errorprov) || is.na(errorprov))errorprov = error1
if (errorprov < error1 && identical(order1,order2)) {
error1 = errorprov
paramprov=coef(nls.out)
} else if (errorprov > worsterror) {
worsterror = errorprov
}
}}
signals_parameters = paramprov
fitted_signals = signal_fitting(signals_parameters,
Xdata,multiplicities,roof_effect,program_parameters$freq)
bins=c()
for (ind in signals_to_fit) {
sorted_bins=sort(fitted_signals[ind,]/sum(fitted_signals[ind, ]),decreasing=T,index.return=T)
if(length(sorted_bins$x)>0) bins= sorted_bins$ix[1:which.min(abs(cumsum(sorted_bins$x)-0.75))]
}
if (length(bins)==0) bins=seq_along(Ydata)
#Correction of half_bandwidth and j-coupling
iter = 0
error22=error2=error1
errorprov = error1=3000
#Only half_bandwidth and j-coupling will have different lower und upper bounds.
change_indexes=which(seq_along(lb)%%5!=3 & seq_along(lb)%%5!=4 & seq_along(lb)%%5!=0)
lb[change_indexes]=ub[change_indexes]=signals_parameters[change_indexes]
while (iter < 3) {
set.seed(iter);s0 = lb + (ub - lb) * runif(length(ub))
nls.out <-
minpack.lm::nls.lm(
par = s0,
fn = residFun,
observed = Ydata,
xx = Xdata,
multiplicities=multiplicities,
roof_effect=roof_effect,
freq=program_parameters$freq,
lower = lb,
upper = ub,
control = minpack.lm::nls.lm.control(
factor = program_parameters$factor,
maxiter = program_parameters$nls_lm_maxiter,
ftol = program_parameters$ftol,
ptol = program_parameters$ptol
)
)
iter = iter + 1
# #Procedure to calculate the fititng error in all the ROI
#An adapted MSE error is calculated, and the parameters of the optimization with less MSE are stored
errorprov = (sqrt(nls.out$deviance / length(Ydata))) * 100 / (max(Ydata) -
min(Ydata))
if (is.nan(errorprov) || is.na(errorprov)) errorprov = error1
if (errorprov < error1) {
error1 = errorprov
paramprov=coef(nls.out)
} else if (errorprov > worsterror) {
worsterror = errorprov
}
if (error1 < error2) {
error2 = error1
signals_parameters = paramprov
}
}
#If half_bandwidth and j-coup change improves fitting
iterrep = iterrep + 1
#If the fitting seems to be still clearly improvable through the addition of signals
if (iterrep <= fitting_maxiterrep& error22 < (program_parameters$additional_signal_improvement * dummy) &
(error22 > program_parameters$additional_signal_percentage_limit)&length(peaks_xdata$maxtab$pos)>sum(multiplicities[signals_to_fit])) {
# print('Trying to improve initial fit adding peaks')
#I find peaks on the residuals
residual_peaks = tryCatch(peakdet(c(nls.out$fvec[1],diff(nls.out$fvec)), program_parameters$peakdet_minimum*max(1e-10,max(Ydata))),error= function(e){
dummy=list(signals_parameters=signals_parameters,error1=error1)
return(dummy)
})
if (is.null(residual_peaks$maxtab) == F) {
#Preparation of information of where signals of interest are located
dummy=multiplicities[signals_to_fit]%%2
dummy[dummy==0]=2
additional_signal_matrix = matrix(paramprov,nrow(FeaturesMatrix),5,byrow = TRUE)
points_to_avoid = abs(rbind(matrix(Xdata,length(signals_to_fit),length(Xdata),byrow = TRUE) - matrix(
additional_signal_matrix[signals_to_fit, 2] - (additional_signal_matrix[signals_to_fit, 5]/dummy)/program_parameters$freq,length(signals_to_fit),length(Xdata)),
matrix(Xdata,length(signals_to_fit),length(Xdata),byrow = TRUE) - matrix(additional_signal_matrix[signals_to_fit, 2] + (additional_signal_matrix[signals_to_fit, 5]/dummy)/program_parameters$freq,length(signals_to_fit),length(Xdata))))
points_to_avoid = apply(points_to_avoid, 1, which.min)
seq_range = c()
for (i in (-range_ind):range_ind) seq_range = append(seq_range, points_to_avoid - i)
#Finding of posible additional signals to incorporate if there are not in zones where the signals o interest are located
residual_peaks=cbind(residual_peaks$maxtab$pos, residual_peaks$maxtab$val)[residual_peaks$maxtab$pos %in% peaks_bindata$maxtab$pos,,drop=F]
valid_residual_peaks = matrix(NA, 0, 2)
if (nrow(residual_peaks)>0) {
for (i in seq(nrow(residual_peaks))) {
if (any(abs(points_to_avoid - residual_peaks[i, 1]) < range_ind) == F) valid_residual_peaks = rbind(valid_residual_peaks, residual_peaks[i, ])
}
#Selection of more intense additional signals
if (nrow(valid_residual_peaks) > program_parameters$signals_to_add) {
ad = sort(valid_residual_peaks[, 2],decreasing = T,index.return = T)$ix
valid_residual_peaks = valid_residual_peaks[ad[1:min(program_parameters$signals_to_add, length(ad))], , drop = F]
}
#Creation of rows to incorporate to FeaturesMatrix
if (nrow(valid_residual_peaks)>0) {
dummy = t(replicate(nrow(valid_residual_peaks),FeaturesMatrix[1,]))
dummy[, 2] = Ydata[valid_residual_peaks[, 1]]
dummy[, 3] = Xdata[valid_residual_peaks[, 1]] - 0.001
dummy[, 4] = Xdata[valid_residual_peaks[, 1]] + 0.001
dummy[, 5]=min(FeaturesMatrix[,5])
dummy[, 6]=min(FeaturesMatrix[,6])
dummy[, c(9,10,12)] = rep(0, nrow(valid_residual_peaks))
dummy[, 11] = rep(1, nrow(valid_residual_peaks))
FeaturesMatrix = rbind(FeaturesMatrix, dummy)
} else {
iterrep = fitting_maxiterrep +1
}
} else {
iterrep = fitting_maxiterrep +1
}}
} else {
iterrep = fitting_maxiterrep +1
}
}
optim_parameters=list(signals_parameters=signals_parameters,error1=error1)
return(optim_parameters)
}
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.