def match(self, graph1, graph2):
"""
Take two graphs are match them. The two graphs must be AbstractMatrixGraphs
with VertexLists representing the vertices.
:param graph1: A graph object
:param graph2: The second graph object to match
:return permutation: A vector of indices representing the matching of elements of graph1 to graph2
:return distance: The graph distance list [graphDistance, fDistance, fDistanceExact]
"""
# Deal with case where at least one graph is emty
if graph1.size == 0 and graph2.size == 0:
permutation = numpy.array([], numpy.int)
distanceVector = [0, 0, 0]
time = 0
return permutation, distanceVector, time
elif graph1.size == 0 or graph2.size == 0:
if graph1.size == 0:
graph1 = SparseGraph(VertexList(graph2.size, graph2.getVertexList().getNumFeatures()))
else:
graph2 = SparseGraph(VertexList(graph1.size, graph1.getVertexList().getNumFeatures()))
numTempFiles = 5
tempFileNameList = []
for i in range(numTempFiles):
fileObj = tempfile.NamedTemporaryFile(delete=False)
tempFileNameList.append(fileObj.name)
fileObj.close()
configFileName = tempFileNameList[0]
graph1FileName = tempFileNameList[1]
graph2FileName = tempFileNameList[2]
similaritiesFileName = tempFileNameList[3]
outputFileName = tempFileNameList[4]
if self.useWeightM:
W1 = graph1.getWeightMatrix()
W2 = graph2.getWeightMatrix()
else:
W1 = graph1.adjacencyMatrix()
W2 = graph2.adjacencyMatrix()
numpy.savetxt(graph1FileName, W1, fmt="%.5f")
numpy.savetxt(graph2FileName, W2, fmt="%.5f")
# Compute matrix similarities
C = self.vertexSimilarities(graph1, graph2)
numpy.savetxt(similaritiesFileName, C, fmt="%.5f")
# Write config file
configFile = open(configFileName, "w")
configStr = "graph_1=" + graph1FileName + " s\n"
configStr += "graph_2=" + graph2FileName + " s\n"
configStr += "C_matrix=" + similaritiesFileName + " s\n"
configStr += "algo=" + self.algorithm + " s\n"
configStr += "algo_init_sol=" + self.init + " s\n"
configStr += "alpha_ldh=" + str(self.alpha) + " d\n"
configStr += "cdesc_matrix=A c\n"
configStr += "cscore_matrix=A c\n"
configStr += "hungarian_max=10000 d\n"
configStr += "algo_fw_xeps=0.01 d\n"
configStr += "algo_fw_feps=0.01 d\n"
configStr += "dummy_nodes=0 i\n"
configStr += "dummy_nodes_fill=" + str(self.rho) + " d\n"
configStr += "dummy_nodes_c_coef=" + str(self.gamma) + " d\n"
configStr += "qcvqcc_lambda_M=" + str(self.lambdaM) + " d\n"
configStr += "qcvqcc_lambda_min=1e-3 d\n"
configStr += "blast_match=0 i\n"
configStr += "blast_match_proj=0 i\n"
configStr += "exp_out_file=" + outputFileName + " s\n"
configStr += "exp_out_format=Compact Permutation s\n"
configStr += "verbose_mode=0 i\n"
configStr += "verbose_file=cout s\n"
configFile.write(configStr)
configFile.close()
fnull = open(os.devnull, "w")
home = expanduser("~")
argList = [home + "/.local/bin/graphm", configFileName]
subprocess.call(argList, stdout=fnull, stderr=fnull)
fnull.close()
# Next: parse input files
outputFile = open(outputFileName, "r")
line = outputFile.readline()
line = outputFile.readline()
line = outputFile.readline()
line = outputFile.readline()
graphDistance = float(outputFile.readline().split()[2])
fDistance = float(outputFile.readline().split()[2])
fDistanceExact = float(outputFile.readline().split()[2])
time = float(outputFile.readline().split()[1])
line = outputFile.readline()
line = outputFile.readline()
permutation = numpy.zeros(max(graph1.getNumVertices(), graph2.getNumVertices()), numpy.int)
i = 0
for line in outputFile:
permutation[i] = int(line.strip()) - 1
i += 1
# Delete files
os.remove(graph1FileName)
os.remove(graph2FileName)
os.remove(similaritiesFileName)
os.remove(configFileName)
os.remove(outputFileName)
distanceVector = [graphDistance, fDistance, fDistanceExact]
return permutation, distanceVector, time
class CitationIterGenerator(object):
"""
A class to load the high energy physics data and generate an iterator. The
dataset is found in http://snap.stanford.edu/data/cit-HepTh.html
"""
def __init__(self, minGraphSize=500, maxGraphSize=None, dayStep=30):
dataDir = PathDefaults.getDataDir() + "cluster/"
edgesFilename = dataDir + "Cit-HepTh.txt"
dateFilename = dataDir + "Cit-HepTh-dates.txt"
#Note the IDs are integers but can start with zero so we prefix "1" to each ID
edges = []
file = open(edgesFilename, 'r')
file.readline()
file.readline()
file.readline()
file.readline()
for line in file:
(vertex1, sep, vertex2) = line.partition("\t")
vertex1 = vertex1.strip()
vertex2 = vertex2.strip()
edges.append([vertex1, vertex2])
#if vertex1 == vertex2:
# print(vertex1)
file.close()
logging.info("Loaded edge file " + str(edgesFilename) + " with " + str(len(edges)) + " edges")
#Keep an edge graph
graph = DictGraph(False)
graph.addEdges(edges)
logging.info("Created directed citation graph with " + str(graph.getNumEdges()) + " edges and " + str(graph.getNumVertices()) + " vertices")
#Read in the dates articles appear in a dict which used the year and month
#as the key and the value is a list of vertex ids. For each month we include
#all papers uploaded that month and those directed cited by those uploads.
startDate = datetime.date(1990, 1, 1)
file = open(dateFilename, 'r')
file.readline()
numLines = 0
subgraphIds = []
for line in file:
(id, sep, date) = line.partition("\t")
id = id.strip()
date = date.strip()
inputDate = datetime.datetime.strptime(date.strip(), "%Y-%m-%d")
inputDate = inputDate.date()
if graph.vertexExists(id):
tDelta = inputDate - startDate
graph.vertices[id] = tDelta.days
subgraphIds.append(id)
#If a paper cites another, it must have been written before
#the citing paper - enforce this rule.
for neighbour in graph.neighbours(id):
if graph.getVertex(neighbour) == None:
graph.setVertex(neighbour, tDelta.days)
subgraphIds.append(neighbour)
elif tDelta.days < graph.getVertex(neighbour):
graph.setVertex(neighbour, tDelta.days)
numLines += 1
file.close()
subgraphIds = set(subgraphIds)
graph = graph.subgraph(list(subgraphIds))
logging.debug(graph)
logging.info("Loaded date file " + str(dateFilename) + " with " + str(len(subgraphIds)) + " dates and " + str(numLines) + " lines")
W = graph.getSparseWeightMatrix()
W = W + W.T
vList = VertexList(W.shape[0], 1)
vList.setVertices(numpy.array([graph.getVertices(graph.getAllVertexIds())]).T)
#Note: we have 16 self edges and some two-way citations so this graph has fewer edges than the directed one
self.graph = SparseGraph(vList, W=W)
logging.debug(self.graph)
#Now pick the max component
components = self.graph.findConnectedComponents()
self.graph = self.graph.subgraph(components[0])
logging.debug("Largest component graph: " + str(self.graph))
self.minGraphSize = minGraphSize
self.maxGraphSize = maxGraphSize
self.dayStep = dayStep
def getIterator(self):
"""
Return an iterator which outputs the citation graph for each month. Note
that the graphs are undirected but we make them directed.
"""
vertexArray = self.graph.getVertexList().getVertices()
dates = vertexArray[:, 0]
firstVertex = numpy.argmin(dates)
dayList = range(int(numpy.min(dates)), int(numpy.max(dates)), self.dayStep)
dayList.append(numpy.max(dates))
subgraphIndicesList = []
#Generate subgraph indices list
for i in dayList:
subgraphIndices = numpy.nonzero(dates <= i)[0]
#Check subgraphIndices are sorted
subgraphIndices = numpy.sort(subgraphIndices)
currentSubgraph = self.graph.subgraph(subgraphIndices)
compIndices = currentSubgraph.depthFirstSearch(list(subgraphIndices).index(firstVertex))
subgraphIndices = subgraphIndices[compIndices]
if self.maxGraphSize != None and subgraphIndices.shape[0] > self.maxGraphSize:
break
print(subgraphIndices.shape[0])
if subgraphIndices.shape[0] >= self.minGraphSize:
subgraphIndicesList.append(subgraphIndices)
iterator = IncreasingSubgraphListIterator(self.graph, subgraphIndicesList)
return iterator
def match(self, graph1, graph2):
"""
Take two graphs are match them. The two graphs must be AbstractMatrixGraphs
with VertexLists representing the vertices.
:param graph1: A graph object
:param graph2: The second graph object to match
:return permutation: A vector of indices representing the matching of elements of graph1 to graph2
:return distance: The graph distance list [graphDistance, fDistance, fDistanceExact]
"""
#Deal with case where at least one graph is emty
if graph1.size == 0 and graph2.size == 0:
permutation = numpy.array([], numpy.int)
distanceVector = [0, 0, 0]
time = 0
return permutation, distanceVector, time
elif graph1.size == 0 or graph2.size == 0:
if graph1.size == 0:
graph1 = SparseGraph(
VertexList(graph2.size,
graph2.getVertexList().getNumFeatures()))
else:
graph2 = SparseGraph(
VertexList(graph1.size,
graph1.getVertexList().getNumFeatures()))
numTempFiles = 5
tempFileNameList = []
for i in range(numTempFiles):
fileObj = tempfile.NamedTemporaryFile(delete=False)
tempFileNameList.append(fileObj.name)
fileObj.close()
configFileName = tempFileNameList[0]
graph1FileName = tempFileNameList[1]
graph2FileName = tempFileNameList[2]
similaritiesFileName = tempFileNameList[3]
outputFileName = tempFileNameList[4]
if self.useWeightM:
W1 = graph1.getWeightMatrix()
W2 = graph2.getWeightMatrix()
else:
W1 = graph1.adjacencyMatrix()
W2 = graph2.adjacencyMatrix()
numpy.savetxt(graph1FileName, W1, fmt='%.5f')
numpy.savetxt(graph2FileName, W2, fmt='%.5f')
#Compute matrix similarities
C = self.vertexSimilarities(graph1, graph2)
numpy.savetxt(similaritiesFileName, C, fmt='%.5f')
#Write config file
configFile = open(configFileName, 'w')
configStr = "graph_1=" + graph1FileName + " s\n"
configStr += "graph_2=" + graph2FileName + " s\n"
configStr += "C_matrix=" + similaritiesFileName + " s\n"
configStr += "algo=" + self.algorithm + " s\n"
configStr += "algo_init_sol=" + self.init + " s\n"
configStr += "alpha_ldh=" + str(self.alpha) + " d\n"
configStr += "cdesc_matrix=A c\n"
configStr += "cscore_matrix=A c\n"
configStr += "hungarian_max=10000 d\n"
configStr += "algo_fw_xeps=0.01 d\n"
configStr += "algo_fw_feps=0.01 d\n"
configStr += "dummy_nodes=0 i\n"
configStr += "dummy_nodes_fill=" + str(self.rho) + " d\n"
configStr += "dummy_nodes_c_coef=" + str(self.gamma) + " d\n"
configStr += "qcvqcc_lambda_M=" + str(self.lambdaM) + " d\n"
configStr += "qcvqcc_lambda_min=1e-3 d\n"
configStr += "blast_match=0 i\n"
configStr += "blast_match_proj=0 i\n"
configStr += "exp_out_file=" + outputFileName + " s\n"
configStr += "exp_out_format=Compact Permutation s\n"
configStr += "verbose_mode=0 i\n"
configStr += "verbose_file=cout s\n"
configFile.write(configStr)
configFile.close()
fnull = open(os.devnull, 'w')
home = expanduser("~")
argList = [home + "/.local/bin/graphm", configFileName]
subprocess.call(argList, stdout=fnull, stderr=fnull)
fnull.close()
#Next: parse input files
outputFile = open(outputFileName, 'r')
line = outputFile.readline()
line = outputFile.readline()
line = outputFile.readline()
line = outputFile.readline()
graphDistance = float(outputFile.readline().split()[2])
fDistance = float(outputFile.readline().split()[2])
fDistanceExact = float(outputFile.readline().split()[2])
time = float(outputFile.readline().split()[1])
line = outputFile.readline()
line = outputFile.readline()
permutation = numpy.zeros(
max(graph1.getNumVertices(), graph2.getNumVertices()), numpy.int)
i = 0
for line in outputFile:
permutation[i] = int(line.strip()) - 1
i += 1
#Delete files
os.remove(graph1FileName)
os.remove(graph2FileName)
os.remove(similaritiesFileName)
os.remove(configFileName)
os.remove(outputFileName)
distanceVector = [graphDistance, fDistance, fDistanceExact]
return permutation, distanceVector, time
