I have been investigating the use of the Intel Math Kernel Library (MKL) and Intel compilers to produce R binaries on Linux, and have included some benchmarks to compare these with GNU compilers and the reference BLAS and LAPACK. I am making my notes public, as the topic may be of interest to others. Bear in mind that this is a work in progress: R-intel-fedora

# Author Archives: Martyn

# Scotland says no

This is a quick update to yesterday’s post on polling for the referendum on the independence of Scotland from the UK. The final result is 55.3% for No (i.e. staying in the union), 44.7% for Yes. This is consistent with the opinion polls, which predicted a win for No, but the margin is certainly bigger than predicted by the opinion polls, especially in the run-up to the referendum.

The image below from whatscotlandthinks.org shows a “poll of polls” – a running mean of opinion polls – which is a smoothed version of the figure I showed yesterday. The average of the last 6 polls on the day before the referendum gave a prediction of 52% for Yes – only a 4 percentage point difference compared with the 11 point difference seen in the referendum. This vindicates Stephen Fisher’s analysis showing that opinion polls tend to overestimate the desire for constitutional change.

If you ignore the narrowing in the gap that apparently occurred in the last two weeks, the polls seemed to give a pretty good prediction from April to August. So perhaps this wasn’t a last minute surge for Yes, but a “patriotic spiral of silence” from No voters, as Martin Boon suggested. But I’ll leave that question to the psephologists.

# The only poll that counts

The people of Scotland are voting today in a referendum to decide whether Scotland should become an independent country (“Yes”) or remain in the United Kingdom (“No”), ending the 300 year old Treaty of Union. As shown by the poll tracker above (from the website whatscotlandthinks.org) opinion polls have consistently put the No vote ahead of Yes. In recent weeks the gap has narrowed but only two polls have put the Yes vote ahead. It looks like a narrow win for No, but can the polls be trusted? Continue reading

# JAGS 3.4.0 for Mavericks

Thanks to Matt Denwood, we now have a binary distribution of JAGS 3.4.0 for Mac OS X Mavericks. There is a separate binary for older Mac OS X releases (Snow Leopard, Lion, Mountain Lion). Both binary packages can be downloaded from the files page.

[*Update 2014-09-11*] The rjags package for Mavericks is now available from CRAN. To make it work, you need to have the right combination of R and JAGS:

- If you installed the R-3.1.1-mavericks.pkg version of R then you need to install JAGS-Mavericks-3.4.0.dmg
- If you installed the R-3.1.1-snowleopard.pkg version of R then you need to install JAGS-SnowLeopard-3.4.0.dmg

If you get an error about a **missing symbol** then you have the wrong combination of R and JAGS.

*[Update 2014-11-28]* Note that if you are running Mavericks or Yosemite then you cannot assume that you are running the “Mavericks” version of R, as the “Snow Leopard” version also runs on the latest releases of Mac OS X. If you are not sure which R binary package is installed then type

.Platform$pkgType

at the R prompt (note the dot at the beginning). This will either say mac.binary or mac.binary.mavericks. Follow the corresponding link to get the right binary version of JAGS to match your R installation.

If you are still having problems, come to the JAGS forums.

# DSC 2014, Day 2

This will be somewhat shorter summary of the second day of DSC 2014. There was a wider variety of presentations today, but I am only going to discuss those that touched on the main theme of the day, which was reproducibility. Continue reading

# DSC 2014. Day 1

This is a report of the first day of the Directions in Statistical Computing (DSC) conference that took place in Brixen, Italy (See here for an introduction). Performance enhancements were the main theme of the day, covering not just improvements to R itself but alternate language implementations. Continue reading

# DSC 2014, Brixen

This is one of three planned posts on the 2014 Directions in Statistical Computing (DSC) conference held in Brixen, Italy on 26-27 June. This one is just scene setting. Continue reading