R for spatial analysis

We're looking forward to the May edition of the Zurich R User Meetup, sponsored by Wüest Partner / Datahouse.

Schedule:

06.15 pm Doors open
06.30 pm Introduction / Welcome by the organizing team and Daniel Meister, CTO Datahouse
06.40 pm Talks (see below)
ca 07.40-09.00 pm Apéro

"Micro location rating in real estate"
Jacqueline Schweizer, Wüest Partner

In the world of real estate “location” seems to be a key term. Location
has substantial influence when appraising a house or an apartment.
However, there are different perspectives and spatial levels where one
can determine the quality of location. The time and effort needed to
qualify a location as “good” or “bad” have historically been strongly
based on human resources. The evaluators had to travel on the premises, ask
around, research further online, and establish a rating that needed to
be as free of subjective values as possible, but never truly turned out
to be. Wüest Partner was, therefore, looking for a better solution. We
developed a GIS-based model to derive the measurable qualities of every
existing location within Switzerland. New and improved data, but also
increasing computing power allowed us to develop a high-resolution
spatial model in R to simplify and accelerate the process of a real
estate evaluation.

"Speeding R up on computer using embarrassingly parallel and multithreaded BLAS computations"
Andreas Papritz, Department of Environmental Systems Science, ETH Zurich and Statistical Office, City of Zurich

Even cheap computers nowadays have multi-core processors with two or more processing units. Starting from version 2.14.0, the basic R distribution includes the package parallel, which allows R users to easily run independent problems in parallel on computers with multi-core processor, benefitting from shortened computing times. Examples of such “embarrassingly parallel” computations where independent tasks “do not need to communicate with each other” include bootstrapping, cross-validation, replicated analyses of similar data sets, etc. Contributed packages such as foreach and snowfall make “embarrassingly parallel” computations even easier. However, long computing times are sometimes not caused by replicated execution of the same tasks, but are the result of solving “large” problems. In many such problems computations can be accelerated substantially if multi-threaded Basic Linear Algebra Subprograms (BLAS) such as OpenBLAS or Intel’s Math Kernel Library (MKL) are linked to R. Substituting OpenBLAS for the standard BLAS library that ships with R is straightforward for Linux and Mac OS X, and for Windows OS Microsoft’s R Open includes MKL. Using examples of geostatistical data analyses I will show in the presentation how to efficiently run “embarrassingly parallel tasks”, and I shall demonstrate how OpenBLAS substantially speeds up fitting models to geostatistical data.