Skip to content

Non linear regression with complex equation

3 messages · jeff_hawkes, Rolf Turner

#
Hi all,
Is it possible to model a function where the unknown parameter appears both
in the fitted equation AND in the determination of other parameters?  E.g.

y = a^2 + b/2 + k

where a = 2/k  and b = k^2 

and the model needs to determine k?  I know this is a very simple equation
(its just an example), the one I am modelling is much more complicated! 

k appears in the equation which the n.l.r model fits, but it also affects
other parameters in the equation.  Please let me know if you know a way of
achieving this.  I realise it is possible to set up a loop where the
modelled value for k is fed back in to a and b, and the model is run again -
but it seems like there should be a more elegant way within one run of the
model.

Thanks,
Jeff

--
View this message in context: http://r.789695.n4.nabble.com/Non-linear-regression-with-complex-equation-tp4425942p4425942.html
Sent from the R help mailing list archive at Nabble.com.
#
Your question is (completely) ill-posed.  What is your actual
model?  What you have said makes no sense at all as it stands.
(Minimal self-contained example .....)

     cheers,

         Rolf Turner
On 28/02/12 09:25, jeff_hawkes wrote:
#
Apologies for the phrasing of the question.
I've sorted the problem (thanks Bert Gunter) by using the curly brackets {}
as below (using a simplified version of my real model). I hope this helps
someone else!

Jeff
-----------------------
Alpha         ip           X
1     0.7106967  0.3616727 0.006027879
2     2.1678517  5.0615917 0.084359861
3     4.4066250 11.2282945 0.187138242
4     9.8495694 18.0534974 0.300891624
5    27.7247098 29.2064434 0.486774057
6    70.6931430 35.3946092 0.589910153
7   133.1240255 46.0347288 0.767245480
8   214.7851844 49.3811149 0.823018582
9   359.5511036 58.5069583 0.975115972
10  748.1840127 57.3744477 0.956240795
11 2129.9844080 60.0000000 1.000000000
+ X~Alpha/(Alpha+1+k*c/(1+k*Fe1))},start=list(k=1e10))
Formula: X ~ Alpha/(Alpha + 1 + k * c/(1 + k * Fe1))

Parameters:
   Estimate Std. Error t value Pr(>|t|)    
k 3.491e+10  7.190e+09   4.856 0.000665 ***
---
Signif. codes:  0 ?***? 0.001 ?**? 0.01 ?*? 0.05 ?.? 0.1 ? ? 1 

Residual standard error: 0.05589 on 10 degrees of freedom

Number of iterations to convergence: 8 
Achieved convergence tolerance: 2.393e-06 


--
View this message in context: http://r.789695.n4.nabble.com/Non-linear-regression-with-complex-equation-tp4425942p4427617.html
Sent from the R help mailing list archive at Nabble.com.