[R] cmptl_analy.R

[Oscar Linares]

Dear Michael,

One key is adjustment of nls optimizer tolerance. I notice it has to be
higher than usual, but, I recovered your noisy "known" parameter values
with an error of K1 (-7%) and k1 (-6%):

#### Miller problem with Dalgaard modifications
##   Linares 1/22/2004

## Solution 1
nls(noisy ~ lsoda(xstart, time, one.compartment.model, c(K1=K1,
k2=k2))[,2],
          data=C1.lsoda,
          start=list(K1=0.3, k2=0.7),
          control=nls.control(maxiter=50,tol=1),
          trace=T
          )
          
0.3594355 :  0.3 0.7 
0.006118908 :  0.3456684 0.3518834 
0.002828044 :  0.4057868 0.4081200 
Nonlinear regression model
  model:  noisy ~ lsoda(xstart, time, one.compartment.model, c(K1 = K1, 
    k2 = k2))[, 2] 
   data:  C1.lsoda 
       K1        k2 
0.4057868 0.4081200
(-19%)    (-18%)   
 residual sum-of-squares:  0.002828044

## Solution 2 (lower nls optimizer tolerance)
nls(noisy ~ lsoda(xstart, time, one.compartment.model, c(K1=K1,
k2=k2))[,2],
          data=C1.lsoda,
          start=list(K1=0.3, k2=0.7),
          control=nls.control(maxiter=50,tol=0.01),
          trace=T
          )
> nls(noisy ~ lsoda(xstart, time, one.compartment.model, c(K1=K1,
k2=k2))[,2],
+           data=C1.lsoda,
+           start=list(K1=0.3, k2=0.7),
+           control=nls.control(maxiter=50,tol=0.01),
+           trace=T
+           )
0.3594355 :  0.3 0.7 
0.006118908 :  0.3456684 0.3518834 
0.002828044 :  0.4057868 0.4081200 
0.002775223 :  0.4092802 0.4104112 
0.002729966 :  0.4085296 0.4114600 
0.001975772 :  0.4628794 0.4688428 
0.001973953 :  0.4628974 0.4682051 
0.001973698 :  0.4631627 0.4686185 
0.001973422 :  0.4673921 0.4729379 
0.001973413 :  0.4673957 0.4729418 
Error in nls(noisy ~ lsoda(xstart, time, one.compartment.model, c(K1 =
K1,  : 
        step factor 0.000488281 reduced below `minFactor' of 0.000976563

#### Solution 3 (increase nls optimizer tolerance a notch)
nls(noisy ~ lsoda(xstart, time, one.compartment.model, c(K1=K1,
k2=k2))[,2],
          data=C1.lsoda,
          start=list(K1=0.3, k2=0.7),
          control=nls.control(maxiter=50,tol=0.1),
          trace=T
          )
> nls(noisy ~ lsoda(xstart, time, one.compartment.model, c(K1=K1,
k2=k2))[,2],
+           data=C1.lsoda,
+           start=list(K1=0.3, k2=0.7),
+           control=nls.control(maxiter=50,tol=0.1),
+           trace=T
+           )
0.3594355 :  0.3 0.7 
0.006118908 :  0.3456684 0.3518834 
0.002828044 :  0.4057868 0.4081200 
0.002775223 :  0.4092802 0.4104112 
0.002729966 :  0.4085296 0.4114600 
0.001975772 :  0.4628794 0.4688428 
0.001973953 :  0.4628974 0.4682051 
Nonlinear regression model
  model:  noisy ~ lsoda(xstart, time, one.compartment.model, c(K1 = K1, 
    k2 = k2))[, 2] 
   data:  C1.lsoda 
       K1        k2 
0.4628974 0.4682051 
(-7%)     (-6%)
 residual sum-of-squares:  0.001973953 

> errK1=(0.4628974-0.5)/0.5
> errK1
[1] -0.0742052
> errk1=(0.4682051-0.5)/0.5
> errk1
[1] -0.0635898

Oscar
Oscar A. Linares, MD, PhD
Prof. Experimental Medicine
Michigan Diabetes Institute 
olinares at med.umich.edu

Oscar A. Linares, MD, PhD              ///////
Michigan Diabetes Institute S c I S O F T ///=20=03
Molecular Medicine Unit       ______////////=20=04
SciSoft Working Group          \_\_\_\/////
Ann Arbor, MI                   \_\_\_\///
Tel. (734) 637-7997              \_\_\_\/
Fax. (734) 453-7019
***************************************************************************
IMPORTANT CONFIDENTIALITY NOTICE: The documents and/or accom...{{dropped}}