Message-ID: <ad6c3fbe-304c-2c6c-7c19-29e45a9faa78@tu-dresden.de>
Date: 2023-03-14T20:51:56Z
From: Thomas Petzoldt
Subject: [R-sig-dyn-mod] Fitting ODE and DDE models with FME using missing data
In-Reply-To: <21257032.FtyC7LvGez@tux>
I agree completely with Johannes. Observational gaps are no technical
problem for FME, because it is possible to define an own cost function,
or use a "long format" for the data. Examples are found in the package
help files and vignettes or a poster:
https://desolve.r-forge.r-project.org/user2014/examples/FME/
But, the data should describe all features of your process to ensure
identifiability. This is a fundamental scientific (or mathematical)
question and not a technical problem of any package.
Thomas
On 14.03.2023 at 20:09 Johannes Ranke wrote:
> Hi,
>
> In principle, this is possible, you just define the cost function based only on
> the observed variables. But you may run into a lack of identifiability of the
> parameters determining the unobserved variable(s), which may result in non-
> convergence, or very wide confidence intervals for these parameters.
>
> Kind regards,
>
> Johannes
>
>
>
> Am Dienstag, 14. M?rz 2023, 19:59:55 CET schrieb Nicola Gambaro:
>> Dear all,
>>
>> I have to estimate some unknown parameters of an ODE/DDE model. However, we
>> only have access to data of some but not all state variables. Also, some
>> state variable data may have observational gaps in between the time series.
>>
>> Is there a way to at least approximately fit the model with FME? If not,
>> could anyone point me to alternative packages or relevant algorithms?
>>
>> Many thanks,
>> Nicola