I am using the dplyr and broom combination and try to fitting regression models depending on the condition inside of the data groups. Finally I want to extract the regression coefficients by each group.
So far I'm getting the same fitting results for all groups (Each group is separated with letters a:f) . It's the main problem.
library(dplyr)
library(minpack.lm)
library(broom)
direc <- rep(rep(c("North","South"),each=20),times=6)
V <- rep(c(seq(2,40,length.out=20),seq(-2,-40,length.out=20)),times=1)
DQ0 = c(replicate(2, sort(runif(20,0.001,1))))
DQ1 = c(replicate(2, sort(runif(20,0.001,1))))
DQ2 = c(replicate(2, sort(runif(20,0.001,1))))
DQ3 = c(replicate(2, sort(runif(20,0.001,1))))
No = c(replicate(1,rep(letters[1:6],each=40)))
df <- data.frame(direc,V,DQ0,DQ1,DQ2,DQ3,No)
fit conditions can be described as follows;
direc=North and if V<J1 do fitting with the equation exp((-t_pw)/f0*exp(-del1*(1-V/J1)^2)) else if direc=Southand V>J2
do fitting with the same equation. In both case, if V<J1& V>J2 are not satisfied return 1 for each case.
UPDATE
I found that conditional nls can be possible conditional-formula-for-nls with the suggestion in this link.
nls_fit=nlsLM(DQ0~ifelse(df$direc=="North"&V<J1, exp((-t_pw)/f0*exp(-del1*(1-V/J1)^2)),1)*ifelse(df$direc=="South"&V>J2, exp((-t_pw)/f0*exp(-del2*(1-V/J2)^2)),1)
,data=df,start=c(del1=1,J1=15,del2=1,J2=-15),trace=T)
nls_fit
Nonlinear regression model
model: DQ0 ~ ifelse(df$direc == "North" & V < J1, exp((-t_pw)/f0 * exp(-del1 * (1 - V/J1)^2)), 1) * ifelse(df$direc == "South" & V > J2, exp((-t_pw)/f0 * exp(-del2 * (1 - V/J2)^2)), 1)
data: df
del1 J1 del2 J2
1.133 23.541 1.079 -20.528
residual sum-of-squares: 16.93
Number of iterations to convergence: 4
Achieved convergence tolerance: 1.49e-08
On the other hand when I try to fit other columns such as DQ1,DQ2 and DQ3;
I tried nls_fit=nlsLM(df[,3:6]~ifelse(.....
Error in nls.lm(par = start, fn = FCT, jac = jac, control = control, lower = lower, : evaluation of fn function returns non-sensible value!
now the problem came down to multiple column fitting. How can I fit multiple columns DQ0:DQ3 ? I checked how to succinctly write a formula with many variables from a data frame? but couldn't find the solution to use in my data frame.
In addition when I do fitting for DQ0 column inside of its groups
as you can see from the output same Del and J parameters are produced for all groups a:f
df_new<- df%>%
group_by(No)%>%
do(data.frame(model=tidy()))>%
ungroup
df_new
Source: local data frame [24 x 6]
No model.term model.estimate model.std.error model.statistic model.p.value
1 a del1 1.132546 9024.255 1.255002e-04 0.9999000
2 a J1 23.540764 984311.373 2.391597e-05 0.9999809
3 a del2 1.079182 27177.895 3.970809e-05 0.9999684
4 a J2 -20.527520 2362268.839 -8.689748e-06 0.9999931
5 b del1 1.132546 9024.255 1.255002e-04 0.9999000
6 b J1 23.540764 984311.373 2.391597e-05 0.9999809
7 b del2 1.079182 27177.895 3.970809e-05 0.9999684
8 b J2 -20.527520 2362268.839 -8.689748e-06 0.9999931
9 c del1 1.132546 9024.255 1.255002e-04 0.9999000
10 c J1 23.540764 984311.373 2.391597e-05 0.9999809
.. .. ... ... ... ... ...
