This package implements functions related to statistical models of the continuous-time evolution of finite-state random variables along branches of a phylogenetic tree with a known shape, with full or partial observations available only at some nodes of the tree. The distinguishing characteristic of this package is that these functions are computed without accumulating numerical errors when the inputs are exactly represented by double precision floating point numbers.

This collection of functions is not meant as a practical tool for biologists but rather as a reference point for methods development researchers. Neither maximum likelihood nor Bayesian inference is implemented, but some of the implemented functions may be useful in the context of inference. The functions may be much slower than their counterparts in other packages. Because of the numerical guarantees of the function evaluations, failure modes include aborting with an error or hanging forever while the internal working precision is repeatedly doubled.

The code has been tested only on Linux, and the installation requires autotools which should be available as autotools-dev on Linux distributions based on debian.

The package depends on recent versions of these C libraries:

• jansson -- C library for encoding, decoding and manipulating JSON data
• gmp -- The GNU Multiple Precision Arithmetic Library
• flint2 -- Fast Library for Number Theory
• arb -- C library for arbitrary-precision interval arithmetic

Various Python libraries could be required to run tests or to run the examples that involve scripting:

• numpy -- Numerical Python library
• scipy -- Scientific Python library
• pandas -- Python data analysis library
• matplotlib -- Python 2D plotting library
• tabulate -- Pretty-print tabular data in Python
• ete3 -- A Python framework for the analysis and visualization of trees.
• dendropy -- A Python library for phylogenetic computing.

The jq tool may be used for json filtering.

Something like the usual autotools installation commands should work if you are lucky:

$ ./autogen.sh
$ ./configure --prefix=/my/prefix CFLAGS='-march=native -O3 -g -Wall -Wextra'
$ make
$ make check
$ make install

A rudimentary Python interface can be installed using the setup.py script and tested as follows:

$ cd src
src$ python setup.py install --user
src$ cd ..
$ nosetests

These features are available through command-line programs with executable names like arbplf-newton-update which operate via JSON files on stdin and stdout. If the Python extension is installed, the features are available in the arbplf module with function names like arbplf_newton_update using a JSON string API.

• ll -- log likelihood
• deriv -- derivative of log likelihood w.r.t. edge rate coefficients
• hess -- hessian matrix of log likelihood w.r.t. edge rate coefficients
• inv-hess -- inverse of hessian matrix
• marginal -- marginal state distributions at nodes
• dwell -- linear combinations of dwell time expectations in states on edges
• trans -- linear combinations of labeled transition count expectations on edges
• em-update -- update edge rate coefficients using one step of EM
• newton-update -- update edge rate coefficients using one step of Newton's method
• newton-delta -- the difference between the newton update and the current values
• newton-refine -- certify an interior local optimum near an initial guess of edge rate coefficients