Suggestion: reparametrize before using optim

[ghost]

As the model is parametrized currently, delta is a value between 0 and 1 and theta is a positive value. When fitting the model with optim this sometimes leads to non-convergence. This can easily be fixed by transforming the parameters for the function RSS passed to optim, so that each parameter can take any real value. For example, take the logit of delta and the log of theta, and then back-transform before passing the parameters to the other functions inside RSS. This solves the convergence issues.

[FohmAnalys]

Thanks! I will re-parametrise the parameters.

[ghost]

Here is an example of how it could be done: https://github.com/BenjaK/SEIR-model-Stockholm/blob/master/Script/Estimate_SEIR_for_sharing_new_incidence.R (note that the main function has been renamed to Stockholm_SEIR).

[fredlind]

Men nog ska logit vara
logit <- function(p) {
log(p) - log(1 - p)
}

och inte

logit <- function(p) {
log(p) / log(1 - p)
}
?

[ghost]

Whoops! Det stämmer.