def __init__(self):
self.proximal = ProximalDendrite()
self.cells = [Cell() for i in range(CELLS_PER_COLUMN)]
self.neighbors = []
self.overlap = 0
self.active = False
self.activeDutyCycle = 0
self.overlapDutyCycle = 0
self.boost = 1
self.scheduledUpdates = None
def __init__(self): self.proximal = ProximalDendrite(); self.cells = [Cell() for i in range(CELLS_PER_COLUMN)] self.neighbors = [] self.overlap = 0 self.active = False self.activeDutyCycle = 0 self.overlapDutyCycle = 0 self.boost = 1 self.scheduledUpdates = None
class Column:
"""A column in an HTM cortical region"""
def __init__(self):
self.proximal = ProximalDendrite()
self.cells = [Cell() for i in range(CELLS_PER_COLUMN)]
self.neighbors = []
self.overlap = 0
self.active = False
self.activeDutyCycle = 0
self.overlapDutyCycle = 0
self.boost = 1
self.scheduledUpdates = None
def setNeighbors(self, neighbors):
self.neighbors = neighbors
def mapColumnToInput(self, inputBit):
self.proximal.listen(inputBit)
def getOverlap(self):
return self.overlap
def getActiveDutyCycle(self):
return self.activeDutyCycle
def getOutputBits(self):
outputBits = InputVector(0)
for cell in self.cells:
outputBits.appendBit(cell.getOutputBit())
return outputBits
def getLearningCells(self, step):
learningCells = []
for cell in self.cells:
if cell.isLearning(step):
learningCells.append(cell)
return learningCells
def getNumConnectedProximalSynapses(self):
return self.proximal.getNumConnectedSynapses()
def isActive(self):
return self.active
def updateOverlap(self):
self.overlap = self.proximal.overlap(self.boost)
def calculateMinLocalActivity(self, desiredLocalActivity):
neighborOverlap = []
for c in self.neighbors:
neighborOverlap.append(c.getOverlap())
neighborOverlap.sort()
if desiredLocalActivity < len(neighborOverlap):
minLocalActivity = neighborOverlap[desiredLocalActivity]
else:
minLocalActivity = 0
return minLocalActivity
def calculateMinDutyCycle(self):
maxDutyCycle = 0
for c in self.neighbors:
maxDutyCycle = max(maxDutyCycle, c.getActiveDutyCycle())
return 0.01 * maxDutyCycle
def updateActiveState(self, desiredLocalActivity):
#Set active state taking into account inhibition
minLocalActivity = self.calculateMinLocalActivity(desiredLocalActivity)
if (self.overlap > 0 and self.overlap > minLocalActivity):
self.active = True
self.proximal.updatePermanences()
else:
self.active = False
def updateParameters(self):
minDutyCycle = self.calculateMinDutyCycle()
self.updateActiveDutyCycle()
self.updateBoost(minLocalActivity)
self.updateOverlapDutyCycle(minLocalActivity)
def updateActiveDutyCycle(self):
activity = 1 if (self.active) else 0
self.activeDutyCycle = DECAY_RATE * self.activeDutyCycle + (
1 - DECAY_RATE) * activity
def updateOverlapDutyCycle(self, minDutyCycle):
self.overlapDutyCycle = DECAY_RATE * self.overlapDutyCycle + (
1 - DECAY_RATE) * self.overlap
if (self.overlapDutyCycle < minDutyCycle):
self.proximal.increasePermanences()
def updateBoost(self):
minDutyCycle = self.calculateMinDutyCycle()
if (self.activeDutyCycle < minDutyCycle):
self.boost += BOOST_INCREMENT
def bestSegmentScoreCell(self, step):
bestSegmentScore = 0
bestCell = None
for cell in self.cells:
score = cell.bestSegmentScore(step)
if (score > bestSegmentScore):
bestSegmentScore = score
bestCell = cell
return bestCell
def bestSegmentCountCell(self):
minSegmentCount = self.cells[0].getNumSegments()
bestCell = self.cells[0]
for cell in self.cells:
segmentCount = cell.getNumSegments()
if (segmentCount < minSegmentCount):
minSegmentCount = segmentCount
bestCell = cell
return bestCell
#BestMatchingCell(c)
# For the given column, return the cell with the best matching segment (as
# defined above). If no cell has a matching segment, then return the cell with
# the fewest number of segments.
def bestMatchingCell(self, step):
cell = self.bestSegmentScoreCell(step)
if (cell is None):
cell = self.bestSegmentCountCell()
return cell
def advanceTimeStep(self):
for cell in self.cells:
cell.advanceTimeStep()
def updateCellActivity(self, allLearningCells):
for cell in self.cells:
cell.setActive(False)
if (not self.active):
return
buPredicted = False
lcChosen = False
for cell in self.cells:
(cellPredicted, cellLearning) = cell.activationReport()
if (cellPredicted):
buPredicted = True
cell.setActive(True)
if (cellLearning):
cell.setLearning(True)
lcChosen = True
if (not buPredicted):
for cell in self.cells:
cell.setActive(True)
if (not lcChosen):
learningCell = self.bestMatchingCell(PREVIOUS_TIME_STEP)
learningCell.setLearning(True)
learningCell.generateActivityUpdates(True, True, allLearningCells,
PREVIOUS_TIME_STEP)
def updateCellPrediction(self, allPrevLearningCells):
for cell in self.cells:
cell.updatePrediction(allPrevLearningCells, CURRENT_TIME_STEP)
def updateCellLearning(self):
for cell in self.cells:
cell.updateLearning(CURRENT_TIME_STEP)
class Column: """A column in an HTM cortical region""" def __init__(self): self.proximal = ProximalDendrite(); self.cells = [Cell() for i in range(CELLS_PER_COLUMN)] self.neighbors = [] self.overlap = 0 self.active = False self.activeDutyCycle = 0 self.overlapDutyCycle = 0 self.boost = 1 self.scheduledUpdates = None def setNeighbors(self,neighbors): self.neighbors = neighbors def mapColumnToInput(self,inputBit): self.proximal.listen(inputBit) def getOverlap(self): return self.overlap def getActiveDutyCycle(self): return self.activeDutyCycle def getOutputBits(self): outputBits = InputVector(0) for cell in self.cells: outputBits.appendBit(cell.getOutputBit()) return outputBits def getLearningCells(self,step): learningCells = [] for cell in self.cells: if cell.isLearning(step): learningCells.append(cell) return learningCells def getNumConnectedProximalSynapses(self): return self.proximal.getNumConnectedSynapses() def isActive(self): return self.active def updateOverlap(self): self.overlap = self.proximal.overlap(self.boost) def calculateMinLocalActivity(self,desiredLocalActivity): neighborOverlap = [] for c in self.neighbors: neighborOverlap.append(c.getOverlap()) neighborOverlap.sort() if desiredLocalActivity < len(neighborOverlap): minLocalActivity = neighborOverlap[desiredLocalActivity] else: minLocalActivity = 0 return minLocalActivity def calculateMinDutyCycle(self): maxDutyCycle = 0 for c in self.neighbors: maxDutyCycle = max(maxDutyCycle,c.getActiveDutyCycle()) return 0.01 * maxDutyCycle def updateActiveState(self,desiredLocalActivity): #Set active state taking into account inhibition minLocalActivity = self.calculateMinLocalActivity(desiredLocalActivity) if (self.overlap > 0 and self.overlap > minLocalActivity): self.active = True self.proximal.updatePermanences() else: self.active = False def updateParameters(self): minDutyCycle = self.calculateMinDutyCycle() self.updateActiveDutyCycle() self.updateBoost(minLocalActivity) self.updateOverlapDutyCycle(minLocalActivity) def updateActiveDutyCycle(self): activity = 1 if (self.active) else 0 self.activeDutyCycle = DECAY_RATE*self.activeDutyCycle + (1-DECAY_RATE)*activity def updateOverlapDutyCycle(self,minDutyCycle): self.overlapDutyCycle = DECAY_RATE*self.overlapDutyCycle + (1-DECAY_RATE)*self.overlap if (self.overlapDutyCycle < minDutyCycle): self.proximal.increasePermanences() def updateBoost(self): minDutyCycle = self.calculateMinDutyCycle() if (self.activeDutyCycle < minDutyCycle): self.boost += BOOST_INCREMENT def bestSegmentScoreCell(self,step): bestSegmentScore = 0 bestCell = None for cell in self.cells: score = cell.bestSegmentScore(step) if (score > bestSegmentScore): bestSegmentScore = score bestCell = cell return bestCell def bestSegmentCountCell(self): minSegmentCount = self.cells[0].getNumSegments() bestCell = self.cells[0] for cell in self.cells: segmentCount = cell.getNumSegments() if (segmentCount < minSegmentCount): minSegmentCount = segmentCount bestCell = cell return bestCell #BestMatchingCell(c) # For the given column, return the cell with the best matching segment (as # defined above). If no cell has a matching segment, then return the cell with # the fewest number of segments. def bestMatchingCell(self,step): cell = self.bestSegmentScoreCell(step) if (cell is None): cell = self.bestSegmentCountCell() return cell def advanceTimeStep(self): for cell in self.cells: cell.advanceTimeStep() def updateCellActivity(self,allLearningCells): for cell in self.cells: cell.setActive(False) if (not self.active): return buPredicted = False lcChosen = False for cell in self.cells: (cellPredicted,cellLearning) = cell.activationReport() if (cellPredicted): buPredicted = True cell.setActive(True) if (cellLearning): cell.setLearning(True) lcChosen = True if (not buPredicted): for cell in self.cells: cell.setActive(True) if (not lcChosen): learningCell = self.bestMatchingCell(PREVIOUS_TIME_STEP) learningCell.setLearning(True) learningCell.generateActivityUpdates(True,True,allLearningCells,PREVIOUS_TIME_STEP) def updateCellPrediction(self,allPrevLearningCells): for cell in self.cells: cell.updatePrediction(allPrevLearningCells,CURRENT_TIME_STEP) def updateCellLearning(self): for cell in self.cells: cell.updateLearning(CURRENT_TIME_STEP)
