def scrapeIDs(self, seed, size):
if type(size) != int:
raise TypeError
def next(ss, qq):
id = qq.popleft()
if id in ss: return None
node = self.makeID(id)
qq.extend(node.out.keys())
ss.add_node(node)
return id
s = NodeSample()
q = deque([self.getNSID(seed)])
while len(s) < size:
id = next(s, q)
if id is not None:
LOG.info("id sample: %s/%s (added %s)" % (len(s), size, id))
s.build()
return s
def evaluateScheme(self, scheme):
"""
Evaluate the given address scheme against the perfect address scheme
from the complete data of the world.
"""
prune = scheme.copy()
prune.delete_vertices(v.index for v in prune.vs.select(lambda vx: vx[NAA] is None))
ss = NodeSample()
for tag in prune.vs[NID]:
ss.add_node(self.getTagInfo(tag).build_node())
MAX = float("inf")
def dist(arc, graph):
return -log(arc[AAT]*graph.vs[arc.target][NAT]/graph.vs[arc.source][NAT])
def nattr(dist):
return exp(-dist)
# build address scheme of input tags
local = ss.build(complete=False)
assert None not in local.vs[NAT]
local.es[AAT_AD] = [dist(arc, local) for arc in local.es]
path = local.shortest_paths(0, weights=AAT_AD)[0]
graph_prune_arcs(local, [k for k,v in sort_v(enumerate(path))])
local.vs[NAA] = [nattr(d) for d in path]
# build address scheme of n tags from world data, where n = len(input tags)
# OPT LOW this rebuilds the entire graph each time, not optimal, but means
# we can just use already-existing implementation of dijkstra from igraph
sw = NodeSample()
tinfo = self.getTagInfo(prune.vs[0][NID])
sw.add_node(tinfo.build_node())
visit = set([0]) # visited nodes, sw vids
trail = [(0, 0.0)] # trail of visited nodes, ss vids
for i in xrange(0, len(prune.vs)-1): # n-1 because root already added
for rtag in tinfo.rtag.iterkeys():
if rtag not in sw:
sw.add_node(self.getTagInfo(rtag).build_node())
world = sw.build(complete=False)
world.es[AAT_AD] = [dist(arc, world) for arc in world.es]
path = world.shortest_paths(0, weights=AAT_AD)[0]
# get next tag in world addr scheme
npath = [MAX if i in visit else v for i, v in enumerate(path)]
d = min(npath)
index = npath.index(d)
tinfo = self.getTagInfo(world.vs[index][NID])
visit.add(index)
if tinfo.tag not in ss:
trail.append((len(ss), d))
ss.add_node(tinfo.build_node())
else:
trail.append((prune.vs[NID].index(tinfo.tag),d)) # OPT LOW
world = ss.build()
assert len(trail) == len(local.vs)
graph_prune_arcs(world, [vid for vid, dist in trail])
trail = dict(trail)
world.vs[NAA] = [nattr(trail[i]) if i in trail else None for i in xrange(0, len(world.vs))]
return AddrSchemeEval(prune, local, world)