Predictive scoring of Point Processes

[rafa.arc...@gmail.com]

Dear all,

Supposing I have two models:
1) Spatial point process with Gaussian Field
2) Spatial point process without Gaussian Field

Both models have the same integration points (Mesh nodes) but model 1 accounts for spatial autocorrelation with a SPDE term and model 2 not.

What is the best way to show based on a predictive scoring that model 1 leads to better predictions than model 2?

Thank you in advance.

Best regards,

Rafael

[Finn Lindgren]

Hi Rafael,

it depends on what type of prediction you care about. But assuming
the random field is there to capture aspects not captured by the
covariates, any prediction score valid for a point process could be
used.
Normally, one aggregates to some grid and applies a prediction score
to aggregated counts.

Finn

> --
> You received this message because you are subscribed to the Google Groups "R-inla discussion group" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to r-inla-discussion...@googlegroups.com.
> To view this discussion on the web, visit https://groups.google.com/d/msgid/r-inla-discussion-group/ace4ed7c-8714-4cdd-a435-16928a35470dn%40googlegroups.com.



--
Finn Lindgren
email: finn.l...@gmail.com

[rafa.arc...@gmail.com]

Hi Finn,

thanks a lot. Are the conditional predictive ordinates (CPO) valid for point processes? Do you maybe have a source/example where someone does this for point processes?

Best regards,

Rafael

[Finn Lindgren]

No, CPO isn't well defined for point process likelihoods.

Some node on prediction scores (that can be applied to e.g. aggregated
counts from point process data):
https://inlabru-org.github.io/inlabru/articles/prediction_scores.html

Point process residuals can be a useful tool to investigate point
process model predictions:
https://inlabru-org.github.io/inlabru/articles/web/2d_lgcp_residuals.html


Finn

> To view this discussion on the web, visit https://groups.google.com/d/msgid/r-inla-discussion-group/03c69d84-d6d9-4262-829d-bfd1101da98bn%40googlegroups.com.