def totalCost(R): """ prototype, just to get a hint of how good it is.. """ C=0 for n in R.nodes(data=True): C+=R.beta*go.sumWeights(R,n[1]['shortest_path']) C+=go.sumWeights(R,n[1]['second_shortest']) return C/len(R.nodes())
def sumPathsDiff(R,e,storeData=False, add=False):
"""
culculates the difference in the sum of the paths from removing/adding edge e.
May store the new paths as well if storeData==True.
This function is a freaking mess.. clean up..
"""
#if not storeData: R=copy.deepcopy(R) #why was this here? extremely slow.
if len(e)<3: raise Exception('sumPathsDiff needs edge data as well. set data=True so e[2] can be reached')
beta=R.beta
w=R.roadWidth #width of roads
C=0 #cost
origin=R.origin
inf=1e15
eps=1e-8
etpl=tuple(e)
if add: #different orders of things, but otherwise the same.
action1=R.add_edges_from
action2=R.remove_edges_from
#we need to reduce the number of nodes checked... if euc. distance
#to origin is less than half that from e, remove
dtmp=fun.getDistance(e[0], R.origin)
lst=[n for n in R.nodes(data=True) if fun.getDistance(n[0], R.origin)>dtmp*0.5]#really slow, but necessary
#lst=R.nodes(data=True) #old one, replaced to increase speed a bit..
else:
action1=R.remove_edges_from
action2=R.add_edges_from
lst=e[2]['visited_from_node'] #faster than above.
action1([etpl])
if not nx.is_connected(R): #probably remove in this case. Not allowed since graph looses connectivity
action2([etpl])
return inf
routeAfter=nx.algorithms.shortest_paths.weighted.single_source_dijkstra(R, origin)
sumP21={}
for n in lst:
P21=routeAfter[1][n[0]]
P21.reverse()
sumP21[n[0]]=go.sumWeights(R, P21) #need to do this before addding/removing edge again
action2([etpl])
#print "sumpathdiff, visited from ", len(e[2]['visited_from_node']), " nodes"
for nTmp in lst:
if nTmp[0]==R.origin: continue
#print nTmp[1]
P11=nTmp[1]['shortest_path']
P12=nTmp[1]['second_shortest']
old_s1=go.sumWeights(R,P11)
#if add and storeData: print "p12:", add, storeData, P12
old_s2=go.sumWeights(R,P12)
P21=routeAfter[1][nTmp[0]] #used to be try statement here, be aware of exceptions
#P21.reverse() #already reversed above now!
#P21=nx.dijkstra_path(R,nTmp[0], origin) #try single source also..may be faster for a lot of n
if abs(sumP21[nTmp[0]]- old_s1) > eps: C+=beta*(sumP21[nTmp[0]]-old_s1)
action1([etpl])
if C<inf: #road back calculated, now road there.
eTmp=[P21[0], P21[1]] #minus, since reveresed. Look above
eTmp.append(R.get_edge_data(eTmp[0], eTmp[1]))
P21W=go.sumWeights(R,P21) #needs to be calculated before eTmp is removed
R.remove_edges_from([tuple(eTmp)]) #temporary remove to get loop
cycle=go.shortestCycle(R,nTmp[0])
alternative=False
altW=inf
W22=None
if cycle: #if we found one..
altW=P21W+go.sumWeights(R,cycle)
tmp=copy.deepcopy(P21)
tmp.reverse() #want from origin to point
alternative=tmp+cycle
try:
P22=nx.dijkstra_path(R, origin, nTmp[0]) #cannot do this single source
if alternative and go.sumWeights(R,P22)>altW:
P22=alternative
W22=altW
else:
W22=go.sumWeights(R,P22)
except nx.exception.NetworkXNoPath:
if alternative:
P22=alternative
W22=altW
else:
C=inf #this exception is thrown if the node is in a local graph separated from OR
R.add_edges_from([tuple(eTmp)]) #reset road
if C>=inf: break #saves some time, but is ugly
if abs(W22-old_s2)>eps:
C+=W22-old_s2
#if P22 != old_s2: C+=go.sumWeights(R,P22)-old_s2 #old one. wrong, right?
if storeData:
nTmp[1]['new_shortest_path']=P21
nTmp[1]['new_second_shortest']=P22
action2([etpl])
#print "C=", C
if C>=inf: R.add_edges_from([tuple(e)]) #we broke out without adding again..
return C