def _chooseBestSegmentPerCell(self,
connections,
cells,
allMatchingSegments,
potentialOverlaps):
"""
For each specified cell, choose its matching segment with largest number
of active potential synapses. When there's a tie, the first segment wins.
@param connections (SparseMatrixConnections)
@param cells (numpy array)
@param allMatchingSegments (numpy array)
@param potentialOverlaps (numpy array)
@return (numpy array)
One segment per cell
"""
candidateSegments = connections.filterSegmentsByCell(allMatchingSegments, cells)
# Narrow it down to one pair per cell.
onePerCellFilter = np2.argmaxMulti(potentialOverlaps[candidateSegments], connections.mapSegmentsToCells(candidateSegments))
learningSegments = candidateSegments[onePerCellFilter]
return learningSegments
def _sensoryComputeLearningMode(self, anchorInput):
"""
Associate this location with a sensory input. Subsequently, anchorInput will
activate the current location during anchor().
@param anchorInput SDR
A sensory input. This will often come from a feature-location pair layer.
"""
overlaps,potentialOverlaps = self.connections.computeActivityFull(anchorInput, False)
activeSegments = np.flatnonzero(overlaps >= self.activationThreshold)
matchingSegments = np.flatnonzero(potentialOverlaps >= self.learningThreshold)
# Cells with a active segment: reinforce the segment
cellsForActiveSegments = self.connections.mapSegmentsToCells(activeSegments)
learningActiveSegments = activeSegments[np.in1d(cellsForActiveSegments, self.getLearnableCells(), assume_unique=True)]
remainingCells = np.setdiff1d(self.getLearnableCells(), cellsForActiveSegments, assume_unique=True)
# Remaining cells with a matching segment: reinforce the best
# matching segment.
candidateSegments = self.connections.filterSegmentsByCell(matchingSegments, remainingCells)
cellsForCandidateSegments = (self.connections.mapSegmentsToCells(candidateSegments))
candidateSegments = candidateSegments[np.in1d(cellsForCandidateSegments, remainingCells, assume_unique=True)]
onePerCellFilter = np2.argmaxMulti(potentialOverlaps[candidateSegments], cellsForCandidateSegments)
learningMatchingSegments = candidateSegments[onePerCellFilter]
newSegmentCells = np.setdiff1d(remainingCells, cellsForCandidateSegments, assume_unique=True)
for learningSegments in (learningActiveSegments, learningMatchingSegments):
self._learn(learningSegments, anchorInput, potentialOverlaps)
# Remaining cells without a matching segment: grow one.
self._learnOnNewSegments(newSegmentCells, anchorInput)
self.activeSegments = activeSegments
self.sensoryAssociatedCells = self.getLearnableCells()
def _chooseBestSegmentPerColumn(self, connections, matchingCells,
allMatchingSegments, potentialOverlaps):
"""
For all the columns covered by 'matchingCells', choose the column's matching
segment with largest number of active potential synapses. When there's a
tie, the first segment wins.
@param connections (SparseMatrixConnections)
@param matchingCells (numpy array)
@param allMatchingSegments (numpy array)
@param potentialOverlaps (numpy array)
"""
candidateSegments = connections.filterSegmentsByCell(
allMatchingSegments, matchingCells)
# Narrow it down to one segment per column.
cellScores = potentialOverlaps[candidateSegments]
columnsForCandidates = (
connections.mapSegmentsToCells(candidateSegments) //
self.cellsPerColumn)
onePerColumnFilter = np2.argmaxMulti(cellScores, columnsForCandidates)
learningSegments = candidateSegments[onePerColumnFilter]
return learningSegments
