class KohonenLayer(Layer):
def __init__(self, nodes, bias, neighbourhoodType, rows, columns, conscience):
Layer.__init__(self, nodes, bias)
self.neighbourhoodType = neighbourhoodType
self.rows = rows
self.columns = columns
self.conscience = conscience
self.neighbourhoodFunction = NeighbourhoodFucntion()
self.winFrequencies = {}
self.nodesCount = len(self.nodes)
for node in self.nodes :
self.winFrequencies[node] = 1.0 / self.nodesCount
self.stepsCount = 0
def __str__(self):
return Layer.__str__(self) + '- dimension - ' + str(self.rows) + ' x ' + str(self.columns)
def learn(self, coefficient, conscienceCoefficient, neighbourhoodWidth):
maximum = None
for node in self.nodes:
current_value = node.get_value() + conscienceCoefficient * (1.0 - float(self.nodesCount) * self.winFrequencies[node])
if (maximum is None) or (current_value > maximum[0]):
maximum = (current_value, node)
if self.conscience:
for node in self.nodes :
self.winFrequencies[node] = self.winFrequencies[node] * self.stepsCount + 1 if node == maximum[1] else self.winFrequencies[node] * self.stepsCount
self.winFrequencies[node] = self.winFrequencies[node] / (self.stepsCount + 1)
self.stepsCount = self.stepsCount + 1
if self.neighbourhoodType:
winnerIndex = self.nodes.index(maximum[1])
for nodeIndex in range(len(self.nodes)) :
if winnerIndex != nodeIndex :
self.nodes[nodeIndex].learn(coefficient * self.neighbourhoodFunction.calculate(self.distance(nodeIndex, winnerIndex), neighbourhoodWidth))
maximum[1].learn(coefficient)
self.winner = maximum[1]
def distance(self, nodeIndex, winnerIndex):
return abs(self.row(nodeIndex) - self.row(winnerIndex)) + abs(self.column(nodeIndex) - self.column(winnerIndex))
def row(self, index):
return index / self.rows
def column(self, index):
return index % self.rows
def propagate(self, debug=False):
maximum = None
for node in self.nodes:
if maximum is None or maximum.get_value() < node.get_value():
maximum = node
maximum.propagate(debug)
self.bias.propagate()
def __init__(self, nodes, bias, neighbourhoodType, rows, columns, conscience):
Layer.__init__(self, nodes, bias)
self.neighbourhoodType = neighbourhoodType
self.rows = rows
self.columns = columns
self.conscience = conscience
self.neighbourhoodFunction = NeighbourhoodFucntion()
self.winFrequencies = {}
self.nodesCount = len(self.nodes)
for node in self.nodes :
self.winFrequencies[node] = 1.0 / self.nodesCount
self.stepsCount = 0