# [R] nls question

Gabor Grothendieck ggrothendieck at gmail.com
Thu Jul 12 22:03:03 CEST 2012

```On Thu, Jul 12, 2012 at 3:40 PM, Felipe Carrillo
<mazatlanmexico at yahoo.com> wrote:
> I get a different error now:
>>  nls(weight ~ cbind(1, exp(gamma*week)), weightData, start = list(gamma=
>> 0.2), alg = "plinear")
> Error in nls(weight ~ cbind(1, exp(gamma * week)), weightData, start =
> list(gamma = 0.2),  :
>   step factor 0.000488281 reduced below 'minFactor' of 0.000976562
> The help file says: ........When start is missing, a very cheap guess for
> start is tried (if algorithm != "plinear").

Please give a reproducible example by setting the seed.  This
reproducible example converges:

> set.seed(123)
> weight_random <- runif(50,1,24)
> weight <- sort(weight_random)
> weightData <- data.frame(weight,week=1:50)
> nls(weight ~ cbind(1, exp(gamma*week)), weightData, start = list(gamma = 0.2), alg = "plinear")
Nonlinear regression model
model:  weight ~ cbind(1, exp(gamma * week))
data:  weightData
gamma      .lin1      .lin2
1.136e-03 -3.949e+02  3.962e+02
residual sum-of-squares: 9.17

Number of iterations to convergence: 8
Achieved convergence tolerance: 9.581e-06

as does nls with Gauss Newton:

> nls(weight ~ alpha + beta*exp(gamma*week), weightData, start =
+ c(alpha = 0.0, beta = 1, gamma = 0.2)
+ )
Nonlinear regression model
model:  weight ~ alpha + beta * exp(gamma * week)
data:  weightData
alpha       beta      gamma
-3.949e+02  3.961e+02  1.136e-03
residual sum-of-squares: 9.17

Number of iterations to convergence: 48
Achieved convergence tolerance: 2.906e-06

>
> So I removed  'plinear' from the call and got the following:
>  nls(weight ~ cbind(1, exp(gamma*week)), weightData,start =
> list(gamma=0.2),trace=TRUE)

It cannot be specified as if it were a plinear model but then use
Gauss-Newton.  See ?nls

--
Statistics & Software Consulting
GKX Group, GKX Associates Inc.
tel: 1-877-GKX-GROUP
email: ggrothendieck at gmail.com

```