LibBi 1.2.0 released, new stochastic volatility example

posted by Lawrence Murray on 25 Aug 2014

LibBi version 1.2.0 is now available: easier to install, Windows compatible for the first time, and with AVX support for faster performance on recent CPUs.

A new package demonstrating a simple stochastic volatility model is also available.

Changes

New features:

Streamlined installation instructions: Perl module dependencies are now installed automatically, make test checks for these dependencies, and MAGMA is optional.
CPU builds now work on Windows under Cygwin.
CUDA builds now work on Mac OS X.
AVX support added (use --enable-avx).
SMC$^2$ now works with a Kalman filter over the state, not just a particle filter, and has been renamed marginal SIR to reflect this.
Output at observation times can now be disabled (use --without-output-at-obs).
Non-const scalar expressions are now allowed as indexes on the left side of actions.

Changes:

Distribution now contains a compiled version of the manual.
Major refactoring of methods code to reduce replication.

Performance:

Improved performance under --enable-sse.

Fixes:

Fixed the first observation in a time series being zero when --start-time did not match the earliest time in --obs-file.
Fixed CUDA kernel execution configurations for large numbers of variables.
Fixed matrix range being ignored on the right side of actions.