def build_potential_graphs(cycle_mapping):
graphs = {}
base_decider = InequalityDecider()
base_decider.add_cycle_mapping(problem_def, cycle_mapping)
decider = base_decider.freeze()
for result, inputs in problem_def.iteritems():
my_cycles = tuple(map(operator.itemgetter(1), filter(lambda a: a[0]==result, cycle_mapping)))
other_cycles = map(operator.itemgetter(1), filter(lambda a: a[0]!=result, cycle_mapping))
all_states = set(map(lambda s: string.join(s, ""), itertools.permutations(["a","a","b","b","c"])))
other_states = all_states-set(itertools.chain(*my_cycles))
for input in inputs:
graphs[input] = (decider.build_transition_graph(my_cycles, other_states, input), my_cycles[0][0])
return graphs
def cycles_compatible(left, right):
if left[1] == right[1]:
# same cycle, different inputs, this cannot be part of a consistent mapping
return False
elif len(set(cycles[left[1]]).intersection(set(cycles[right[1]]))) == 0:
# disjoint cycles are always locally compatible
return True
elif left[0] == right[0]:
# this is an adjoint cycle, it can only be compatible if the intersection of states is empty
# therefore it is not compatible
return False
else:
# there is some overlap, we need to evaluate pairwise consistency explicitly
decider = InequalityDecider()
decider.add_cycle(problem_def, left[0], cycles[left[1]])
decider.add_cycle(problem_def, right[0], cycles[right[1]])
return decider.satisfiable()
def build_decider(prefix):
basedecider = InequalityDecider()
basedecider.add_cycle_mapping(problem_def, cycle_mapping)
for transition in prefix:
basedecider.add_transition(*transition)
return basedecider.freeze()
other_cycles = map(operator.itemgetter(1), filter(lambda a: a[0]!=result, cycle_mapping))
all_states = set(map(lambda s: string.join(s, ""), itertools.permutations(["a","a","b","b","c"])))
other_states = all_states-set(itertools.chain(*my_cycles))
for input in inputs:
graphs[input] = (decider.build_transition_graph(my_cycles, other_states, input), my_cycles[0][0])
return graphs
result = list()
for i in xrange(len(cliques)):
if i % total_slices == my_slice:
clique = cliques[i]
print "starting %d %s"%(i, str(clique))
cycle_mapping = map(lambda c: (c[0], tuple(cycles[c[1]])), clique)
decider = InequalityDecider()
decider.add_cycle_mapping(problem_def, cycle_mapping)
if decider.satisfiable():
print " satisfiable"
graphs = build_potential_graphs(cycle_mapping)
good_states = list()
for state in set(map(lambda s: string.join(s, ""), itertools.permutations(["a","a","b","b","c"]))):
try:
decider = InequalityDecider()
decider.add_cycle_mapping(problem_def, cycle_mapping)
for input, graph_plus in graphs.iteritems():
path = nx.shortest_path(graph_plus[0], state, graph_plus[1])
for f, t in zip(path[:-1], path[1:]):
decider.add_transition(f, t, input)
# use list/set to remove duplicates
return reversed(sorted(list(set([clique] + children)), key=len))
output = set()
total_slices = 399
basedir = sys.argv[2]
my_slice = int(sys.argv[4])
for vertex in itertools.product(problem_def.keys(), xrange(len(cycles))):
if index(vertex) % total_slices == my_slice:
graph = nx.read_gpickle("%s/pruned_graphs/%d-%d.gpickle.bz2"%(basedir, vertex[0], vertex[1]))
print "processing %s (%d nodes)"%(str(vertex), len(graph))
for clique in subgraph_cliques(graph, vertex, valid_decoding):
print " is valid %s"%str(clique)
consistents = set()
for subclique in subcombine(clique, vertex):
# we dont need to process a subclique in case a super-clique has been found to be consistent
if not any(subclique < c for c in consistents):
cycle_mapping = map(lambda c: (c[0], cycles[c[1]]), subclique)
decider = InequalityDecider()
decider.add_cycle_mapping(problem_def, cycle_mapping)
if decider.satisfiable():
print " is consistent %s"%str(subclique)
consistents.add(subclique)
output.add(tuple(sorted(subclique, key=index)))
with open("%s/cliques/%d.pickle"%(basedir, my_slice), "w") as f:
pickle.dump(list(output), f)
