domino/interactive.py

IMP::domino::DominoSampler supports an interactive mode where by each step of the sampling process is called explicitly and all intermediate results are exposed. This usage mode can be used to help understand how domino behaves as well as to distribute a domino computation to multiple cores or multiple machines. For the latter, it is useful to use the IMP::domino::get_assignments() and IMP::domino::set_assignments() functions to save the states to a file.

## \example domino/interactive.py
# IMP::domino::DominoSampler supports an interactive mode where by each step of the sampling process is called explicitly and all intermediate results are exposed. This usage mode can be used to help understand how domino behaves as well as to distribute a domino computation to multiple cores or multiple machines. For the latter, it is useful to use the IMP::domino::get_assignments() and IMP::domino::set_assignments() functions to save the states to a file.
#

import IMP.domino
import IMP.algebra
import IMP.container
import IMP

m = IMP.kernel.Model()

# create some particles
ps = [IMP.core.XYZ.setup_particle(IMP.kernel.Particle(m)) for x in range(0, 3)]

s = IMP.core.HarmonicDistancePairScore(1, 1)
lpc = IMP.container.ListPairContainer(
    [(ps[i[0]], ps[i[1]]) for i in [(0, 1), (1, 2)]])
print [(p[0].get_name(), p[1].get_name()) for p in lpc.get_particle_pairs()]
r = IMP.container.PairsRestraint(s, lpc)
r.set_maximum_score(.1)

space = IMP.domino.XYZStates(
    [IMP.algebra.Vector3D(i, 0, 0) for i in range(0, 6)])

pst = IMP.domino.ParticleStatesTable()
for p in ps:
    pst.set_particle_states(p, space)

m.set_log_level(IMP.base.SILENT)

# make sure to break up the
mt = IMP.domino.get_merge_tree([r], pst)
try:
    IMP.show_graphviz(mt)
except:
    print "Unable to display graph using 'dot'"

ds = IMP.domino.DominoSampler(m, pst)
# use the default setup for filters
ds.set_scoring_function([r])
ds.set_merge_tree(mt)
ds.set_log_level(IMP.base.SILENT)

# recurse down the tree getting the assignments and printing them


def get_assignments(vertex):
    on = mt.get_out_neighbors(vertex)
    if len(on) == 0:
        # we have a leaf
        ret = ds.get_vertex_assignments(vertex)
    else:
        # recurse on the two children
        if on[0] > on[1]:
            # the get_vertex_assignment methods expects the children in sorted
            # order
            on = [on[1], on[0]]
        a0 = get_assignments(on[0])
        a1 = get_assignments(on[1])
        ret = ds.get_vertex_assignments(vertex, a0, a1)
    print mt.get_vertex_name(vertex), [str(r) for r in ret]
    return ret

# the root is the last vetex
get_assignments(mt.get_vertices()[-1])


schedule = []
# we could instead decompose the tree into independent sets of jobs


def schedule_job(vertex):
    # first figure out the maximum phase for my children, then add me to the
    # next higher phase
    on = mt.get_out_neighbors(vertex)
    max_child_time = -1
    for n in on:
        max_child_time = max(schedule_job(n), max_child_time)
    my_time = max_child_time + 1
    while len(schedule) < my_time + 1:
        schedule.append([])
    # add myself to the next free phase
    schedule[my_time].append(vertex)
    return my_time

schedule_job(mt.get_vertices()[-1])
print "The merging can be scheduled as", schedule