Computing transformations on `ppp` objects

[mjemons]

Dear authors

I am trying to write a small package that can extract variables from a SpatialExperiment object, transform them into a point pattern object and then compute point process models on them

A minimal example below shows a current limitation I face. Ideally, I would like to be able to specify functions/transformations on ppp objects with the R formula interface. However, mpl.get_covariates does not let me pass ppp objects in data to ppm. Could you maybe relax this assertion to allow for transformations of ppp objects which in turn are then windows or images? Precomputing would work too, but is less elegant in my opinion. Some functions like e.g. s(x) do work as being passed directly as transformations, only the functions on pppobjects pose a problem.

Thank you in advance for your help,
Martin

library("spatstat.model")
ama <- amacrine[square(0.7)]
resp <- spatstat.geom::unmark(ama[ama$marks %in% "on", drop = TRUE])
off <- spatstat.geom::unmark(ama[ama$marks %in% "off", drop = TRUE])

data <- list(response = resp,
            off = off)

ppm(response ~ distfun(off), data = data)
## fails

data[["doff"]] <- with(ama,distfun(off))

ppm(response ~ doff, data = data)
## works

[baddstats]

This is not a simple matter.

ppm converts the point process model into a generalised linear model. ppm interprets the data and evaluates each covariate at the quadrature points, producing a data frame of covariate values. This data frame is passed to glm which fits the model using essentially the same model formula. The formula response ~ distfun(off) does not fit into this scheme, because it will cause glm to look for a column of data named off and a function called distfun that maps real numbers to real numbers.

To do what you want, we would have to partially evaluate terms on the right-hand side of the formula. The sub-expression distfun(off) would be recognised as needing to be evaluated, then would be dispatched to be evaluated, yielding a distfun object, and this function would then be evaluated at the quadrature points. Partially evaluating the right hand side of the formula would be complicated and would break some of the design decisions.

I will think about possible strategies, but again, this is not a simple matter.

[mjemons]

Dear Adrian,

Thank you for the fast response. I realised that this was already not too easy to be solved on my end. I tried out pre-computing and changing the formula-environment, all with limited success. This is why I thought it might be easier to solve this in spatstat directly.

I can try to formalise my problem better and then write a pre-computing wrapper on my end if this is not a current development priority for spatstat.

Best and thanks again
Martin