I have N experimental data sets where I am fitting a complicated non-linear model (where I need to solve a lot of mathematical equations) and this model includes 6 fitting parameters. I have fitted the model to the experimental values, however, apparently the parameters set is not unique and each data set produces different values for the fitting parameters. I want to generate a "predictive" model in the end, so I am looking for a methodology, with which I can calculate the "average" values, or a set of fitting parameters that I can use it globally with a minimized global error.
Hi. What about fitting all of the data sets with one regression model that has a dummy variable for which data set the observation came from. I expect this would work. Best, David
Thank you for your response David:
I am aware of such solutions, but there is two problems: (1) I need to solve so many sophisticated mathematical equations that is not feasible to fit my parameters all at once in terms of computational cost; (2) my model has multiple responses to the optimization process and by combining all the data sets together, I make it even worse.
So, by now, I have good fitting results (different in fitted parameter values thought), but want to know if there is any logical statistical method to find a "common" data set out of the fitted parameters.
I believe that there is no "common" approach other than the one I suggested. BTW if it takes a week of computer time if the project is important I would do it. How do you know that combining data sets makes the problem worse? Have you tried it? My advice: Research does not proceed by finding reasons not to do things but rather to do new things. Best, David
- Badges
- Science topic
- Similar topics
- Mathematics