exabayes 1.4 released

294 views
Skip to first unread message

Andre J. Aberer

unread,
Aug 27, 2014, 4:20:17 AM8/27/14
to exab...@googlegroups.com
Dear ExaBayes users,

Recently, ExaBayes has been accepted for publication in "Molecular
Biology and Evolution"
http://mbe.oxfordjournals.org/content/early/2014/08/26/molbev.msu236.full

We would like to celebrate this by releasing ExaBayes, version 1.4, that
comes with several improvements:

* improved branch length sampling: in collaboration with Fredrik
Ronquist, we developed a new Newton-Raphson-based branch length
proposal with substantial improvements in sampling efficiency
(blDistGamma). It replaces the node slider and branch length
multiplier.

* exabayes now by default starts from a parsimony tree.

* improved guided SPR: exabayes-1.4 introduces a posterior-guided SPR
move (likeSPR). This move is particularly expensive to compute and
thus by default has a low weight (i.e., is not used very
often). However, we found that it noticeably reduces burn-in times and
increases mixing efficiency. Thus w.r.t. to runtime, exabayes-1.4
computes less generations per runtime than its predecessors. We
recommend to specifically use this move (and tune its parameters) when
dealing with hard-to-resolve many-taxon phylogenies.

* Important: we added a new section to the manual about how to trade
runtime versus convergence behavior. This is specifically relevant for
most efficient usage of the guided proposals, such as the likeSPR or
parsimony-guided SPR.

* exabayes now compiles with icc-14.

* two optional parameters are introduced that allows you to use hybrid
proposals (proposing topology and branch lengths
simultaneously). While we found hybrid proposals to improve
convergence in some cases, the overall benefit was limited. If you
want to try out these options, search the manual for *useMultipler*
and *moveOptMode*.

* If you are not satisfied with the performance of exabayes-1.4 on your
dataset, you easily can switch back to the proposal configuration of
exabayes-1.4 and modify the proposal block as follows:

[ topology ]
eTBR 5
eSPR 5
stNNI 5
parsimonySPR 5
likeSpr 0

[ branch lengths ]
branchMulti 15
treeLengthMult 1
nodeSlider 5
blDistGamma 0
biasBLMult 0

--
Best regards,
Andre

Christopher Hann-Soden

unread,
Jan 8, 2016, 5:31:23 PM1/8/16
to ExaBayes
Hi Andre,

The documentation only describes how to compile with GCC or Clang. I am having difficulty installing with icc-14.0.4. Is compiling with icc-14 documented anywhere?

Thanks!
Christopher

Andre Aberer

unread,
Jan 9, 2016, 9:44:50 AM1/9/16
to exab...@googlegroups.com
Hi Christopher,

ICC is quite tough. Up until recently, I did not deem it possible, since
C++11 support in icc is problematic.

However, with the latest devel version, I managed to compile both the
MPI and threaded version using icc-14.

Most likely, you'll have to configure exabayes to use icc/icpc. Thus,

./configure CC=icc CXX=icpc

Since icc is a compiler only, you'll also need a c++ standard library
(typically, provided by gcc). In other words, if you try to compile it
on a cluster and the cluster has a module system, then make sure that
you have a gcc-module loaded (apart from the icc-14 and possibly a MPI
module). In my case, loading the icc-14 and gcc-4.8 module worked
(gcc-4.9 did not work though).

As said, you need to use the very latest development version.
https://github.com/aberer/exabayes
(make sure it says branch devel)

For convenience, here is a link to a tar-ball that is based on commit
a922648:
http://aberer.io/software/exabayes-devel.tar.gz

--
Best regards,
Andre
--
Sent with my mu4e

Christopher Hann-Soden

unread,
Jan 21, 2016, 6:32:24 PM1/21/16
to ExaBayes, an...@aberer.io
Hi Andre,

Using this tar-ball, gcc-4.8.5, and icc-14 with open-MPI 1.6.5 seems to have worked. Thank you so much for your help!

Best,
Christopher
Reply all
Reply to author
Forward
0 new messages