Hi all, I’ve been trying to use the methods from this paper to estimate spatial segregation between crashes involving pedestrians versus those not in Indianapolis. The authors made this R package to facilitate the necessary workflow. However, the size of the dataset makes it such that performing the monte carlo analysis in R takes hours to complete (and it’s supposed to performed more than once in a somewhat iterative process). I’m fairly confident it would be faster in Julia. Is there currently a means to replicate the same workflow in Julia? Would be happy to provide more information if you can help guide me with what you need. Thanks!
I’ve been looking for this type of stuff in Julia myself. I doubt it exists.
I’m actually interested in submitting code for Conley (1999) spatially correlated standard errors. But first I’d like to see a more organization in the Julia stats ecosystem. Right now it’s a bit messy.
The link to the R package seems to be wrong, since it is the same as the paper.
Perhaps @juliohm knows if this is possible using existing tools?
Thanks! I fixed the link.
Anything that can be done in R can be done in Julia. Someone just needs to do it :). You can even call R from Julia. Rather then wait for someone else to clean up the ecosystem, make a package, share code - it’s best to take the bull by the horns and write the code yourself :).
People will help you in your efforts, and you’ll learn way more this way.
Thank you @visr for pinging.
@danriggins I didn’t have a chance to read carefully the paper, but from the package description they are using some kernel density estimator for the intensity of the point process, is that correct?
Please keep an eye on the project. Contributions are very welcome.
@juliohm, yes they use the estimator to create intensity fields for point processes then use monte carlo analysis to identify where the intensities of each point process significantly segregate from each other.
I didn’t read the paper, but if you can assume that there is no spatial interaction, you can simulate a Poisson process with estimated intensity and reject the hypothesis. Currently, only homogeneous (constant intensity) Poisson processes are implemented in PointPatterns.jl, however, and before we can implement inhomogeneous sampling, we need to solve a major issue in the GeoStats.jl stack regarding volumetric measurements, which is on the top of my TODO list. The plan is to include point processes with interaction as well, and increasingly match the features available in R packages.