def isoBFS(G):
table = {}
isoTool.setGraph(G)
todo = [ (0,0,[]) ]
best = bestKnown
bestList = []
hitTotal = 0
def end(signal=None, func=None):
print
print "Considered:", hitTotal
print best
print '\n'.join(str(x) for x in bestList)
print hitCount
if signal:
import sys
sys.exit()
import signal
signal.signal(signal.SIGINT, end)
sortCount = 0
# Main BFS loop
while len(todo) > 0:
hitTotal += 1
#print todo
(cost, depth, cons) = heappop(todo)
#(cost, depth, cons) = todo.pop()
#print todo
#print c
#print triList
#verb = False
#if getConsList(cons) == ['m', 'm', '1bec f08dccc 707af4d', '[]']:
# print "yep"
# print cons
# print calcSize(cons)
# verb = True
# #exit()
#print "cons:", c
# Add in all of the edges needed for G
c = cons
while c!= []:
ta = c[1]
addEdge(G, (ta[1],ta[2]))
c = c[3]
#th = isoTool.getHash() # doesn't have a performance impact
th = 1
#print "considering:", cons
currentSize, usedSet = calcSize(cons)
#if best != -1 and calcSize(cons)[0] > best:
if best != -1 and currentSize > best:
# strip off the edges added at this stage of the BFS
c = cons
while c!= []:
ta = c[1]
delEdge(G, (ta[1],ta[2]))
c = c[3]
continue
edgeCache = {}
# loop across all of the tripples to check where to continue the search
for t in getAllTripples(G):
e1 = sortPair([t[0], t[2]])
e2 = sortPair([t[1], t[2]])
#print "tripple:", t, e1, e2
#print "calling left:"
if edgeCache.has_key(e1):
h1, leafList1 = edgeCache[e1]
else:
addEdge(G,e1)
h1 = isoTool.getHash()
leafList1 = extraLeafCheck(G,topDepth,e1)
#leaf1, cons1 = leafCheck(G)
edgeCache[e1] = h1, leafList1
delEdge(G,e1)
#print "calling right:"
if edgeCache.has_key(e2):
h2, leafList2 = edgeCache[e2]
else:
addEdge(G,e2)
h2 = isoTool.getHash()
#if h1 != h2:
leafList2 = extraLeafCheck(G,topDepth,e2)
#leaf2, cons2 = leafCheck(G)
edgeCache[e2] = h2, leafList2
delEdge(G,e2)
#print "leafList1:", leafList1
#print "leafList2:", leafList2
#for leaf1, cons1 in leafList1:
# for leaf2, cons2 in leafList2:
for leafNode1 in leafList1:
for leafNode2 in leafList2:
haveCons = False
#if h1 == h2 or (leafNode1[0] and leafNode1[0] == leafNode2[0]):
if h1 == h2:
leftBranch = ('m', [], [], [])
#if leafNode1[0] in ('4', '5', 'h'):
if leafNode1[0]:
rightBranch = leafNode1
haveCons = True
h = h1, h2
#h = False
else:
rightBranch = []
h = h1
# uhm. Not sure about this. Seems like a big change:
#nt = t[:]
nt = t[1], t[0], t[2]
elif leafNode1[0] and leafNode2[0]:
leftBranch = leafNode1
rightBranch = leafNode2
nt = t[:]
haveCons = True
h = sortPair((h1,h2))
#h = False
elif leafNode1[0]:
leftBranch = leafNode1
rightBranch = []
nt = t[:]
h = h2
elif leafNode2[0]:
leftBranch = leafNode2
rightBranch = []
nt = t[1], t[0], t[2]
h = h1
else:
continue
newCons = insertBranch(cons,[False, nt, leftBranch, rightBranch])
#newSize, used = calcSize(newCons)
newSize, used = calcSize([False, nt, leftBranch, rightBranch], usedSet)
newSize = newSize + currentSize - 1
#if newSize2+currentSize-1 != newSize or used != used2:
# print "fail"
# print cons
# print newCons
# print [False, nt, leftBranch, rightBranch]
# print
# print currentSize, usedSet
# print newSize, used
# print newSize2, used2
# exit()
score = newSize
#score = 1.0/newSize
#score = (depth+1)/(newSize**2*1.0)
#score = (newSize**2*1.0)/(depth+1)
#print newCons
#print getConsList(newCons)
#print "result:", (h,used,newSize)
#print "in table:", (h,used,newSize) in table
if (h,used,newSize) not in table:
table[(h,used,newSize)] = True
print "remaining:", len(todo), "best:", best, "size:", newSize, "score:", score, "cons:", getConsList(newCons)
if haveCons: # we have a construction
if best == -1 or newSize < best:
best = newSize
bestList = []
if newSize == best:
#bestList.append((c, t, code, (leaf1, leaf2), (cons1, cons2),used))
bestList.append(newCons)
elif best == -1 or newSize <= best:
heappush(todo, (score,depth+1,newCons))
#todo.append((score,depth+1,newCons))
#todo.insert(0, (score,depth+1,newCons))
# strip off the edges added at this stage of the BFS
c = cons
while c!= []:
ta = c[1]
delEdge(G, (ta[1],ta[2]))
c = c[3]
#table[th] = True
end()
def isoBFS(G):
table = {}
isoTool.setGraph(G)
todo = [(0, 0, [])]
best = bestKnown
bestList = []
hitTotal = 0
def end(signal=None, func=None):
print
print "Considered:", hitTotal
print best
print '\n'.join(str(x) for x in bestList)
print hitCount
if signal:
import sys
sys.exit()
import signal
signal.signal(signal.SIGINT, end)
sortCount = 0
# Main BFS loop
while len(todo) > 0:
hitTotal += 1
#print todo
(cost, depth, cons) = heappop(todo)
#(cost, depth, cons) = todo.pop()
#print todo
#print c
#print triList
#verb = False
#if getConsList(cons) == ['m', 'm', '1bec f08dccc 707af4d', '[]']:
# print "yep"
# print cons
# print calcSize(cons)
# verb = True
# #exit()
#print "cons:", c
# Add in all of the edges needed for G
c = cons
while c != []:
ta = c[1]
addEdge(G, (ta[1], ta[2]))
c = c[3]
#th = isoTool.getHash() # doesn't have a performance impact
th = 1
#print "considering:", cons
currentSize, usedSet = calcSize(cons)
#if best != -1 and calcSize(cons)[0] > best:
if best != -1 and currentSize > best:
# strip off the edges added at this stage of the BFS
c = cons
while c != []:
ta = c[1]
delEdge(G, (ta[1], ta[2]))
c = c[3]
continue
edgeCache = {}
# loop across all of the tripples to check where to continue the search
for t in getAllTripples(G):
e1 = sortPair([t[0], t[2]])
e2 = sortPair([t[1], t[2]])
#print "tripple:", t, e1, e2
#print "calling left:"
if edgeCache.has_key(e1):
h1, leafList1 = edgeCache[e1]
else:
addEdge(G, e1)
h1 = isoTool.getHash()
leafList1 = extraLeafCheck(G, topDepth, e1)
#leaf1, cons1 = leafCheck(G)
edgeCache[e1] = h1, leafList1
delEdge(G, e1)
#print "calling right:"
if edgeCache.has_key(e2):
h2, leafList2 = edgeCache[e2]
else:
addEdge(G, e2)
h2 = isoTool.getHash()
#if h1 != h2:
leafList2 = extraLeafCheck(G, topDepth, e2)
#leaf2, cons2 = leafCheck(G)
edgeCache[e2] = h2, leafList2
delEdge(G, e2)
#print "leafList1:", leafList1
#print "leafList2:", leafList2
#for leaf1, cons1 in leafList1:
# for leaf2, cons2 in leafList2:
for leafNode1 in leafList1:
for leafNode2 in leafList2:
haveCons = False
#if h1 == h2 or (leafNode1[0] and leafNode1[0] == leafNode2[0]):
if h1 == h2:
leftBranch = ('m', [], [], [])
#if leafNode1[0] in ('4', '5', 'h'):
if leafNode1[0]:
rightBranch = leafNode1
haveCons = True
h = h1, h2
#h = False
else:
rightBranch = []
h = h1
# uhm. Not sure about this. Seems like a big change:
#nt = t[:]
nt = t[1], t[0], t[2]
elif leafNode1[0] and leafNode2[0]:
leftBranch = leafNode1
rightBranch = leafNode2
nt = t[:]
haveCons = True
h = sortPair((h1, h2))
#h = False
elif leafNode1[0]:
leftBranch = leafNode1
rightBranch = []
nt = t[:]
h = h2
elif leafNode2[0]:
leftBranch = leafNode2
rightBranch = []
nt = t[1], t[0], t[2]
h = h1
else:
continue
newCons = insertBranch(
cons, [False, nt, leftBranch, rightBranch])
#newSize, used = calcSize(newCons)
newSize, used = calcSize(
[False, nt, leftBranch, rightBranch], usedSet)
newSize = newSize + currentSize - 1
#if newSize2+currentSize-1 != newSize or used != used2:
# print "fail"
# print cons
# print newCons
# print [False, nt, leftBranch, rightBranch]
# print
# print currentSize, usedSet
# print newSize, used
# print newSize2, used2
# exit()
score = newSize
#score = 1.0/newSize
#score = (depth+1)/(newSize**2*1.0)
#score = (newSize**2*1.0)/(depth+1)
#print newCons
#print getConsList(newCons)
#print "result:", (h,used,newSize)
#print "in table:", (h,used,newSize) in table
if (h, used, newSize) not in table:
table[(h, used, newSize)] = True
print "remaining:", len(
todo
), "best:", best, "size:", newSize, "score:", score, "cons:", getConsList(
newCons)
if haveCons: # we have a construction
if best == -1 or newSize < best:
best = newSize
bestList = []
if newSize == best:
#bestList.append((c, t, code, (leaf1, leaf2), (cons1, cons2),used))
bestList.append(newCons)
elif best == -1 or newSize <= best:
heappush(todo, (score, depth + 1, newCons))
#todo.append((score,depth+1,newCons))
#todo.insert(0, (score,depth+1,newCons))
# strip off the edges added at this stage of the BFS
c = cons
while c != []:
ta = c[1]
delEdge(G, (ta[1], ta[2]))
c = c[3]
#table[th] = True
end()
def extraLeafCheck(G, depth, ne = None, nKiteEdges=None):
global hitCount
# do a depth limited search for a construction.
# So we take both edges of a triple, and both of the resulting graphs are leafs
# depth is how many more times we can recurse before finding a leaf.
# bigTable is a global cache of graphs and depths, but we can only cache negative results.
# caching on graph isomorphism here doesn't work.
# You cache a result. Then you pull that result from the cache later for a new graph.
# But the indicies are wrong, so your construction is wrong when you sum the graphs.
# To make this work, you would have to map back and forth between
# the canonical indicies and your working indices.
#h1 = -1
# depth is part of the hash because just because a graph isn't a leaf at depth 4, doesn't mean it's not a leaf at depth 3
h1 = isoTool.getHash()
if (h1, depth) in bigTable:
# print "Cache hit:", h1, bigTable[h1]
hitCount += 1
return bigTable[(h1, depth)]
# test if this graph is a leaf. I.E. contains a 4-clique, 5-wheel, or h graph
l = leafCheck(G,ne,nKiteEdges)
# return if we have a leaf, or we have hit our depth limit.
if l[0] or depth <= 2: # leafCheck takes care of depth <= 2
#bigTable[h1] = [l] # FIXME I think we can cache the negative result here for depth
return [l]
# edgeCache caches a result for when an edge participates in multiple tripples.
edgeCache = {}
retList = [] # we need to return all possible solutions
usedSets = [] # a list of sets of graphs used for each of the solutions in retLest
kiteEdges = getKiteEdges(G) # performance optimization
for t in getAllTripples(G):
e1 = sortPair([t[0], t[2]])
e2 = sortPair([t[1], t[2]])
#print "elc tripple:", t, e1, e2
# Basic recursive check on the edges
if edgeCache.has_key(e1):
r1 = edgeCache[e1]
else:
addEdge(G,e1)
r1 = extraLeafCheck(G,depth-1,e1,kiteEdges)
delEdge(G,e1)
edgeCache[e1] = r1
if edgeCache.has_key(e2):
r2 = edgeCache[e2]
else:
addEdge(G,e2)
r2 = extraLeafCheck(G,depth-1,e2,kiteEdges)
delEdge(G,e2)
edgeCache[e2] = r2
# to make this arbitrary depth:
# leaf1 and leaf2 become lists
# is that it?
# Oh, and we are going to need edges for all returns
# loop across the results
for l1 in r1:
for l2 in r2:
# Found a pair of results that are good.
# non-good results are in the ret-list? We don't just return empty?
if l1[0] and l2[0]:
#print "t:", t
#print "derp1:", l1
#print "derp2:", l2
#print G
#print "big1:", r1
#print "big2:", r2
#exit()
# get the subgraph generated by the hajos sum of the leaf graphs
# r is presumably our standard search node quad. Whatever that is.
# this reduces our found construction to a subgraph construction
r = getCombinedCons(G, t, l1, l2)
# might be able to optimize the above. We always call it, so we might
# not need to recurse. In fact, it's kind of goofy that this gets called recursively.
#if r[0] == '1bec e0328 4f4d838':
# print r
# exit()
# Do some optimzation to reduce the size of our results
# use calcSize to get the set of graphs used in the construction
used = calcSize(r)[1]
# calcSize doesn't include the root graph in the set, so add it
used = used.union([r[0]])
# Check if we have already seen a result that uses a subset of the
# graphs that we are using.
okay = True
for s in usedSets:
if s.issubset(used):
okay = False
break
# now the other side of the above. Go through the list, and kill any previous results that
# are supersets of the current result.
if okay:
#this is going to be hacky.
#filter the retList if the new construction is a subset of
# any of the constructions in the retlist.
newUsed = []
newRet = []
for i in xrange(len(retList)):
s = usedSets[i]
if not used.issubset(s):
newUsed.append(s)
newRet.append(retList[i])
retList = newRet
usedSets = newUsed
retList.append(r)
usedSets.append(used)
#print "elc ret:", retList
if retList:
return retList # we have a result, we can return it.
# unfortunately, our search node format does not cover returning usedSets as well.
else:
bigTable[(h1,depth)] = [(False,)] # only cache on failure
return [(False,)]
def extraLeafCheck(G, depth, ne=None, nKiteEdges=None):
global hitCount
# do a depth limited search for a construction.
# So we take both edges of a triple, and both of the resulting graphs are leafs
# depth is how many more times we can recurse before finding a leaf.
# bigTable is a global cache of graphs and depths, but we can only cache negative results.
# caching on graph isomorphism here doesn't work.
# You cache a result. Then you pull that result from the cache later for a new graph.
# But the indicies are wrong, so your construction is wrong when you sum the graphs.
# To make this work, you would have to map back and forth between
# the canonical indicies and your working indices.
#h1 = -1
# depth is part of the hash because just because a graph isn't a leaf at depth 4, doesn't mean it's not a leaf at depth 3
h1 = isoTool.getHash()
if (h1, depth) in bigTable:
# print "Cache hit:", h1, bigTable[h1]
hitCount += 1
return bigTable[(h1, depth)]
# test if this graph is a leaf. I.E. contains a 4-clique, 5-wheel, or h graph
l = leafCheck(G, ne, nKiteEdges)
# return if we have a leaf, or we have hit our depth limit.
if l[0] or depth <= 2: # leafCheck takes care of depth <= 2
#bigTable[h1] = [l] # FIXME I think we can cache the negative result here for depth
return [l]
# edgeCache caches a result for when an edge participates in multiple tripples.
edgeCache = {}
retList = [] # we need to return all possible solutions
usedSets = [
] # a list of sets of graphs used for each of the solutions in retLest
kiteEdges = getKiteEdges(G) # performance optimization
for t in getAllTripples(G):
e1 = sortPair([t[0], t[2]])
e2 = sortPair([t[1], t[2]])
#print "elc tripple:", t, e1, e2
# Basic recursive check on the edges
if edgeCache.has_key(e1):
r1 = edgeCache[e1]
else:
addEdge(G, e1)
r1 = extraLeafCheck(G, depth - 1, e1, kiteEdges)
delEdge(G, e1)
edgeCache[e1] = r1
if edgeCache.has_key(e2):
r2 = edgeCache[e2]
else:
addEdge(G, e2)
r2 = extraLeafCheck(G, depth - 1, e2, kiteEdges)
delEdge(G, e2)
edgeCache[e2] = r2
# to make this arbitrary depth:
# leaf1 and leaf2 become lists
# is that it?
# Oh, and we are going to need edges for all returns
# loop across the results
for l1 in r1:
for l2 in r2:
# Found a pair of results that are good.
# non-good results are in the ret-list? We don't just return empty?
if l1[0] and l2[0]:
#print "t:", t
#print "derp1:", l1
#print "derp2:", l2
#print G
#print "big1:", r1
#print "big2:", r2
#exit()
# get the subgraph generated by the hajos sum of the leaf graphs
# r is presumably our standard search node quad. Whatever that is.
# this reduces our found construction to a subgraph construction
r = getCombinedCons(G, t, l1, l2)
# might be able to optimize the above. We always call it, so we might
# not need to recurse. In fact, it's kind of goofy that this gets called recursively.
#if r[0] == '1bec e0328 4f4d838':
# print r
# exit()
# Do some optimzation to reduce the size of our results
# use calcSize to get the set of graphs used in the construction
used = calcSize(r)[1]
# calcSize doesn't include the root graph in the set, so add it
used = used.union([r[0]])
# Check if we have already seen a result that uses a subset of the
# graphs that we are using.
okay = True
for s in usedSets:
if s.issubset(used):
okay = False
break
# now the other side of the above. Go through the list, and kill any previous results that
# are supersets of the current result.
if okay:
#this is going to be hacky.
#filter the retList if the new construction is a subset of
# any of the constructions in the retlist.
newUsed = []
newRet = []
for i in xrange(len(retList)):
s = usedSets[i]
if not used.issubset(s):
newUsed.append(s)
newRet.append(retList[i])
retList = newRet
usedSets = newUsed
retList.append(r)
usedSets.append(used)
#print "elc ret:", retList
if retList:
return retList # we have a result, we can return it.
# unfortunately, our search node format does not cover returning usedSets as well.
else:
bigTable[(h1, depth)] = [(False, )] # only cache on failure
return [(False, )]