def __init__(self, examplesFileName):
"""
Create the class by reading examples from a Matlab file. Instantiate the SVM
and create a preprocesor to standarise examples to have zero mean and unit variance.
"""
self.examplesList = ExamplesList.readFromFile(examplesFileName)
self.examplesList.setDefaultExamplesName("X")
self.examplesList.setLabelsName("y")
(freqs, items) = Util.histogram(self.examplesList.getSampledDataField("y").ravel())
logging.info("Distribution of labels: " + str((freqs, items)))
logging.info("The base error rate is " + str(float(min(freqs))/self.examplesList.getNumExamples()))
self.classifier = LibSVM()
self.errorMethod = Evaluator.balancedError
self.preprocessor = Standardiser()
X = self.preprocessor.standardiseArray(self.examplesList.getDataField(self.examplesList.getDefaultExamplesName()))
self.examplesList.overwriteDataField(self.examplesList.getDefaultExamplesName(), X)
def graphFromMatFile(matFileName):
"""
Generate a sparse graph from a Matlab file of ego and alters and their transmissions. This is a mostly
disconnected graph made up of pairs of connected vertices, i.e each vertex has degree 1.
"""
examplesList = ExamplesList.readFromMatFile(matFileName)
numExamples = examplesList.getNumExamples()
numFeatures = examplesList.getDataFieldSize("X", 1)
numVertexFeatures = numFeatures/2+1
vList = VertexList(numExamples*2, int(numVertexFeatures))
sGraph = SparseGraph(vList)
for i in range(0, examplesList.getNumExamples()):
v1Index = i*2
v2Index = i*2+1
example = examplesList.getSubDataField("X", numpy.array([i])).ravel()
vertex1 = numpy.r_[example[0:numFeatures/2], numpy.array([1])]
vertex2 = numpy.r_[example[numFeatures/2:numFeatures], numpy.array([0])]
sGraph.setVertex(v1Index, vertex1)
sGraph.setVertex(v2Index, vertex2)
sGraph.addEdge(v1Index, v2Index)
return sGraph
def fullTransGraph(self):
"""
This function will return a new graph which contains a directed edge if
a transmission will occur between two vertices.
"""
if self.iteration != 0:
raise ValueError("Must run fullTransGraph before advanceGraph")
#First, find all the edges in the graph and create an ExampleList
numEdges = self.edges.shape[0]
X = numpy.zeros((numEdges*2, self.numPersonFeatures*2))
ind = 0
for i in range(numEdges):
vertex1 = self.graph.getVertex(self.edges[i,0])
vertex2 = self.graph.getVertex(self.edges[i,1])
X[ind, :] = numpy.r_[vertex1[0:self.numPersonFeatures], vertex2[0:self.numPersonFeatures]]
X[ind+numEdges, :] = numpy.r_[vertex2[0:self.numPersonFeatures], vertex1[0:self.numPersonFeatures]]
ind = ind + 1
name = "X"
examplesList = ExamplesList(X.shape[0])
examplesList.addDataField(name, X)
examplesList.setDefaultExamplesName(name)
if self.preprocessor != None:
X = self.preprocessor.process(examplesList.getDataField(examplesList.getDefaultExamplesName()))
examplesList.overwriteDataField(examplesList.getDefaultExamplesName(), X)
y = self.egoPairClassifier.classify(examplesList.getSampledDataField(name))
fullTransmissionGraph = SparseGraph(self.graph.getVertexList(), False)
transIndices = numpy.nonzero(y==1)[0]
#Now, write out the transmission graph
for i in range(len(transIndices)):
if transIndices[i] < numEdges:
fullTransmissionGraph.addEdge(self.edges[transIndices[i],0], self.edges[transIndices[i],1])
else:
fullTransmissionGraph.addEdge(self.edges[transIndices[i]-numEdges,1], self.edges[transIndices[i]-numEdges,0])
return fullTransmissionGraph
def advanceGraph(self):
#First, find all the edges in the graph and create an ExampleList
blockSize = 5000
X = numpy.zeros((blockSize, self.numPersonFeatures*2))
possibleTransmissionEdges = []
possibleTransmissionEdgeIndices = []
for i in range(self.edges.shape[0]):
vertex1 = self.graph.getVertex(self.edges[i,0])
vertex2 = self.graph.getVertex(self.edges[i,1])
if vertex1[self.infoIndex] == 1 and vertex2[self.infoIndex] == 0:
X[len(possibleTransmissionEdges), :] = numpy.r_[vertex1[0:self.numPersonFeatures], vertex2[0:self.numPersonFeatures]]
possibleTransmissionEdges.append((self.edges[i,0], self.edges[i,1]))
possibleTransmissionEdgeIndices.append(i)
if vertex2[self.infoIndex] == 1 and vertex1[self.infoIndex] == 0:
X[len(possibleTransmissionEdges), :] = numpy.r_[vertex2[0:self.numPersonFeatures], vertex1[0:self.numPersonFeatures]]
possibleTransmissionEdges.append((self.edges[i,1], self.edges[i,0]))
possibleTransmissionEdgeIndices.append(i)
#Increase X if it is small
if (len(possibleTransmissionEdges) == X.shape[0]):
X = numpy.r_[X, numpy.zeros((blockSize, self.numPersonFeatures*2))]
#Now, remove from edges the ones that can possible have a transmission
self.edges = numpy.delete(self.edges, possibleTransmissionEdgeIndices, 0)
X = X[0:len(possibleTransmissionEdges), :]
name = "X"
examplesList = ExamplesList(X.shape[0])
examplesList.addDataField(name, X)
examplesList.setDefaultExamplesName(name)
if self.preprocessor != None:
X = examplesList.getDataField(examplesList.getDefaultExamplesName())
X = self.preprocessor.standardiseArray(X)
examplesList.overwriteDataField(examplesList.getDefaultExamplesName(), X)
y = self.egoPairClassifier.classify(X)
transmissionEdges = numpy.zeros((sum(y==1), 2), numpy.int)
j = 0
#Now, update the vertices to reflect transfer
for i in range(len(possibleTransmissionEdges)):
if y[i] == 1:
transmissionEdges[j, 0] = possibleTransmissionEdges[i][0]
transmissionEdges[j, 1] = possibleTransmissionEdges[i][1]
self.transmissionGraph.setVertex(int(transmissionEdges[j, 0]), self.graph.getVertex(transmissionEdges[j, 0]))
self.transmissionGraph.setVertex(int(transmissionEdges[j, 1]), self.graph.getVertex(transmissionEdges[j, 1]))
self.transmissionGraph.addEdge(int(transmissionEdges[j, 0]), int(transmissionEdges[j, 1]), 1)
j += 1
vertex = self.graph.getVertex(possibleTransmissionEdges[i][1])
vertex[self.infoIndex] = 1
self.graph.setVertex(possibleTransmissionEdges[i][1], vertex)
self.allTransmissionEdges.append(transmissionEdges)
self.iteration = self.iteration + 1
return self.graph
