Changes in 1.8.17
 * [BUGFIX] Make sure proposals and priors are on the same scale.
 * [BUGFIX] We MUST sample both nodes and alignment in SPR moves.
 * [NEW]    Add some documentation.
 * [NEW]    Make tool 'tree-dist-compare' compare any number of 
            tree distributions.
 * [CHANGE] Increase mixing for epsilon...
 * [CHANGE] Sanitize the interface.
 * [CHANGE] Make the makefile self-contained.
 * [CHANGE] Introduce shift_laplace prior to allow data to overpower
            prior.  Use in prior on kappa + omega.

Changes in 1.8.13:
 * [BUGFIX] Our branch length proposal was biased upwards! Fixed.
 * [CHANGE] Sanitize move attributes
 * [CHANGE] Make 'gaps=star' automatically disable conflicting moves.
 * [CHANGE] Print out weights
 * [DEBUG]  In debug mode, check that our cached transition matrices
            are correct.

Changes in 1.8.12:
 * [BUGFIX] Make tree operator=(*this) do nothing.
 * Store partial emission probabilities in Matrix instead of vector<
   valarray<double> >.
 * Compute and cache emission probabilities e(i,j)
 * Alter s2_sub to reduce the number of operations needed to compute
   e(i,j)

Changes in 1.8.11:
 * Compute GQ probabilities using new State-Reduction (SR) algorithm
 * Add new routine logdiff(x,y) = log(exp(x)-exp(y))
 * Unify several routines to just call sample_two_nodes_multi()
 * Unify several routines to just call sample_tri_multi()

Changes in 1.8.9:
 * Change the name 'sampler' to 'bali-phy'.
 * [CHANGE] Further improve and unify loading of alignments w/ and w/o
            internal nodes.
 * [CHANGE] Update alignment-blame and make it run again
 * [CHANGE] Check that target matches samples!
 * Put debugging output from optimize.C under #ifndef NDEBUG
 * [CHANGE] Don't complain about missing letters while trying to
            determine the alphabet.
 * [BUGFIX] Fix tree_sample::supports_topology when topology NEVER
   occurs

Changes in 1.8.8: 
 * [FEATURE] Change randomize_alignment to be useable with large
             number of taxa.
 * [BUGFIX] Change topology branch score by factor of 2
 * [CHANGE] Make alignments print only every 10 iterations to decrease
            file size.
 * [CHANGE] Improve and unify loading of alignments w/ and w/o
            internal nodes
 * [CHANGE] Check that internal node states are not letters, and are
            consistent when loading alignments.
 * [CHANGE] Change mode of kappa distribution to 2, and increase
            variance.
 * [CHANGE] Change kappa initial value to 2.
 * [CHANGE] Improve tree-dist-compare
 * move calculations of likelihoods on half-trees to alignment-sums.C
 * remove move slide_branch_and_T
 * break off part of class HMM into class DPengine
 * break off part of hmm.H into dp-engine.H
 * remove TreeFunc - it wasn't really used

Changes in 1.8.7:
 * [FEATURE] Throw an exception if we set a parameters twice on the
             command line 
 * [FEATURE] New tool tree-dist-compare.C, new file tree-dist.C
 * [BUGFIX]  We accidentally changed SPR_and_A to SPR_and_nodes in
             1.8.6. Change it back! 
 * [BUGFIX] Correctly print the branch length for the highest numbered
            branch.
 * [BUGFIX] The "branch score" distance was broken

Changes in 1.8.6
 * [BUGFIX] remember to recalc() the branches that SPR changes!
 * [CHANGE] Change branch_mean proposals to log scale
 * [CHANGE] Change kappa,omega proposals to log scale in HKY and
            YangCodonModel.


Changes in 1.8.5:
 * [BUGFIX] Con't crash when a MCMC MoveEach has no enabled moves
            for an arg.  Instead, assert.
 * [CHANGE] Run the tree moves twice every iteration....
 * [CHANGE] Run SPR more often