[SciPy-User] Fitting procedure to take advantage of cluster

[Giovanni Luca Ciampaglia]

Il 30. 06. 11 00:46, Jose Gomez-Dans ha scritto:
>
> We looked at Gaussian Proces emulators too, as Giovanni suggested (see 
> the papers by O'Hagan too). However, the problem is that our model 
> typically has several outputs (think of it as correlated time series, 
> for example, a time series of the outflow of rivers in a basin). This 
> isn't easy to do with GPs. However, if your model provides a scalar, 
> then they can be very efficient and are easy to implement.

Hi Jose,
You might want to have a look at the paper by Dancik et al. 
(http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2789658/) where they use 
dimensionality reduction (e.g. PCA) in order to use GP with a model 
whose output is a time series. In my case I had to fit a model whose 
output was a whole population, i.e. a distributional output, so in the 
end I used an auxiliary model (in particular a mixture of gaussians) to 
fit the output of my simulations, and then GPs to learn the mapping 
between the parameters of my model and the sufficient statistic of the 
mixture of gaussians. At that point you can define an error function and 
solve a minimization problem to get the parameter estimates. A bit 
complicated but it works.

Cheers,

-- 
Giovanni Luca Ciampaglia

Ph.D. Candidate
Faculty of Informatics
University of Lugano
Web: http://www.inf.usi.ch/phd/ciampaglia/

Bertastraße 36 ∙ 8003 Zürich ∙ Switzerland