Markku Karhunen wrote:
> Thank you all.
>
> We must think about implementing these packages. In the meantime, I
> should clarify my question: Is there any evidence that doing the dumb
> for loop discretisation is any more dangerous in R, than in any other
> language? Apparently not?
I know of no such evidence. However, I've found that using a standard
package where feasible often brings substantial benefits, e.g., plots &
support functions I might not otherwise have found.
>
> Best,
> Markku Karhunen
>> have you looked at lsoda{odesolve}?
>>
>> have you looked at the scripts\CSTR subdirectory in the fda package?
>> it includes an example worked in both R and Matlab with slightly
>> better answers in R but with a much longer compute time.
>>
>> sg
>>
>> The fda package
>>
>> Peter Dalgaard wrote:
>>> Markku Karhunen wrote:
>>>
>>>> Thanks, Dr. Maechler.
>>>>
>>>>> No, there's no such track.
>>>>> [ Matlab users coming to R may produce wrong R code
>>>>> by using 0:n-1 instead of 0:(n-1) ; but I don't assume this
>>>>> would be the case ]
>>>>>
>>>>>
>>>> Been there, done that!
>>>>
>>>>> MK> We use just a simple discretisation written in a for loop
>>>>> MK> and a variable (i.e. user-fed) time step.
>>>>>
>>>>> I don't think you should use your own code instead of "professional"
>>>>> ODE solvers, such as the one in R package 'odesolve'....
>>>>>
>>>>>
>>>> We must look into that. The problem, maybe, is that in fact half of
>>>> the equations are, in fact, simple PDE's and I don't know, if you
>>>> can put them into odesolve.
>>>>
>>> Usually, you can convert them to a system of ODE's ("the method of
>>> lines" if i remember correctly).
>>>
>>> One slight caveat with the high-end ODE solvers is that sometimes they
>>> are too smart for their own good when used in connection with parameter
>>> estimation. Because of things like adaptive stepsizing, you might
>>> end up
>>> with sums of squared residuals that are non-smooth functions of the
>>> parameters. This happens especially easily if the system itself is not
>>> quite smooth (e.g. if your input to the system is a step function).
>>>
>>>
>>>>> MK> Maybe, I'm too neurotic about this, but I guess I just
>>>>> want some comfort MK> after seeing a few particularly nasty
>>>>> orbits.
>>>>>
>>>>> As we know ``from Chaos theory'', there can be delicate
>>>>> inhereent and numerical problems in ODE solving..
>>>>>
>>>> But - to our best knowledge - they should not be any more acute in
>>>> R, than on any other platform...
>>>>
>>>> BR,
>>>> Markku
>>>>
>>>> ______________________________________________
>>>> R-help@r-project.org mailing list
>>>> https://stat.ethz.ch/mailman/listinfo/r-help
>>>> PLEASE do read the posting guide
>>>> http://www.R-project.org/posting-guide.html
>>>> and provide commented, minimal, self-contained, reproducible code.
>>>>
>>>
>>>
>>>
>
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
