Notes from the Kölner R meeting, 26 June 2015

Last Friday the Cologne R user group came together for the 14th time. For the first time we met at Startplatz, a start-up incubator venue. The venue was excellent, not only did they provide us with a much larger room, but also with table-football and drinks. Many thanks to Kirill for organising all of this!

Photo: Günter Faes

We had two excellent advanced talks. Both were very informative and well presented.

Data Science at the Command Line

Kirill Pomogajko showed us how he uses various command line tools to pre-process log-files for further analysis with R.

Photo: Günter Faes

Imagine you have several servers that generate large data sets with no standard delimiters, like the example below.

	Race Age Military Color Speed Car Month State
	Other 24 Navy Aquamarine4 63 "Lotus Europa" June Michigan
	White 26 Army Palegreen4 53 "Merc 450SE" Florida
	White 26 Army Burlywood3 45 "Porsche 914-2" May Wisconsin
	Hispanic 28 Navy Burlywood3 56 "Mazda RX4 Wag" April Florida
	White 22 Marine Corps Salmon1 51 "Hornet 4 Drive" California
	White 31 Army Lightyellow2 46 "Cadillac Fleetwood" Arizona
	Black 23 Air Force Gray27 58 "Datsun 710" October South Carolina
	White 22 Navy Violetred4 53 "Toyota Corona" California
	White 31 Marine Corps Sienna4 48 "Toyota Corolla" December Oklahoma
	White 24 Navy Rosybrown4 59 "Merc 280C" March New Jersey
	White 27 Army Rosybrown4 61 "Pontiac Firebird" Florida
	Hispanic 20 Army Lightyellow2 37 "Mazda RX4 Wag" July Pennsylvania
	Hispanic 32 Navy Lightyellow2 63 "Volvo 142E" Michigan
	Hispanic 34 Navy Sienna4 36 "Merc 280C September" Nevada
	Hispanic 29 Air Force Aquamarine4 56 "Toyota Corona" Mississippi
	White 28 Air Force Lightyellow2 73 "Honda Civic" November West Virginia
	Asian 26 Army Aquamarine4 64 "Fiat X1-9" March Missouri
	White 23 Army Rosybrown4 53 "Duster 360" May Tennessee
	White 28 Marine Corps Palegreen4 52 "Chrysler Imperial" California

view raw MessyData.txt hosted with ❤ by GitHub

The columns appear to be separated by a blank at first glance, but the second column (Military) has strings such as Air Force that include a blank itself. Furthermore, other columns have missing data (Month) and another uses speech-marks (Car). Thus, it's messy and difficult to read into R.

To solve the problem Kirill developed a Makefile that uses tools such as scp, sed and awk to download and clean the server files.

Kirill's tutorial files are available via GitHub.

An Introduction to RStan and the Stan Modelling Language

Paul Viefers gave a great introduction to Stan and RStan, with a focus on explaining the differences to other MCMC packages such as JAGS.

Photo: Günter Faes

Stan is a probabilistic programming language for Bayesian inference. One of the major challenges in Bayesian analysis is that often there is no analytical solution for the posterior distribution. Hence, the posterior distribution is approximated via simulations, such as Gibbs sampling in JAGS. Stan, on the other hand, uses Hamiltonian Monte Carlo (HMC), an algorithm that is more subtle in proposing jumps, using more structure by translation into Hamiltonian mechanics framework.

Paul ended his talk by walking us through the various building blocks of a Stan script, using a hierarchical logistic regression example.

You can access Paul's slides on RPubs.

Drinks and Networking

No Cologne R user group meeting is complete without Kölsch and networking. In the end some of us ended up in a fancy burger place.

Next Kölner R meeting

The next meeting will be scheduled in September. Details will be published on our Meetup site. Thanks again to Revolution Analytics for their sponsorship.

Posted by Unknown