[R] KDE on a large dataset (15 million points)

Maurizio GIBIN maurizio.gibin at jrc.ec.europa.eu
Thu Apr 9 13:11:14 CEST 2015

Hello everyone,
I am having some troubles in calculating a kde surface from a datatable.
The code is efficient in my opionion, but I have a problem on the output.
After some modelling that is not relevant for the purpose of my question 
I obtain a data.table of around 15 million records (for half a year, I 
forecast around 50 millions for the entire year).
The data.table is pretty simple, just lat and lon.
Let's forget for a second the problems related to projection, and focus 
on the K(ernel)D(ensity)E(stimation). Do you have any suggestion on a 
library that could be able to cope with so many points?
I know I could calculate the density in different ways and through 
different platforms, however, I would like to stick to R as it is 
commonly diffused in research.
I also know that probably subsetting the dataset is a wise choice...


*Scientific/Technical Project Officer

**European Commission**
*Joint Research Centre

Institute for the Protection and Security of the Citizen (IPSC)

Maritime Affairs Unit*

Via Enrico Fermi 2749*
*I-21027 Ispra (VA)*
*+39 0332 786770*
maurizio.gibin at jrc.ec.europa.eu*

**<mailto:maurizio.gibin at jrc.ec.europa.eu> *

More information about the R-help mailing list