def getRFCenters(sp, params, type='connected'):
numColumns = np.product(sp.getColumnDimensions())
dimensions = (params['nX'], params['nY'])
meanCoordinates = np.zeros((numColumns, 2))
avgDistToCenter = np.zeros((numColumns, 2))
for columnIndex in range(numColumns):
receptiveField = np.zeros((sp.getNumInputs(), ))
if type == 'connected':
sp.getConnectedSynapses(columnIndex, receptiveField)
elif type == 'potential':
sp.getPotential(columnIndex, receptiveField)
else:
raise RuntimeError('unknown RF type')
connectedSynapseIndex = np.where(receptiveField)[0]
if len(connectedSynapseIndex) == 0:
continue
coordinates = []
for synapseIndex in connectedSynapseIndex:
coordinate = coordinatesFromIndex(synapseIndex, dimensions)
coordinates.append(coordinate)
coordinates = np.array(coordinates)
coordinates = coordinates.astype('float32')
angularCoordinates = np.array(coordinates)
angularCoordinates[:, 0] = coordinates[:, 0] / params['nX'] * 2 * np.pi
angularCoordinates[:, 1] = coordinates[:, 1] / params['nY'] * 2 * np.pi
for i in range(2):
meanCoordinate = np.arctan2(
np.sum(np.sin(angularCoordinates[:, i])),
np.sum(np.cos(angularCoordinates[:, i])))
if meanCoordinate < 0:
meanCoordinate += 2 * np.pi
dist2Mean = angularCoordinates[:, i] - meanCoordinate
dist2Mean = np.arctan2(np.sin(dist2Mean), np.cos(dist2Mean))
dist2Mean = np.max(np.abs(dist2Mean))
meanCoordinate *= dimensions[i] / (2 * np.pi)
dist2Mean *= dimensions[i] / (2 * np.pi)
avgDistToCenter[columnIndex, i] = dist2Mean
meanCoordinates[columnIndex, i] = meanCoordinate
return meanCoordinates, avgDistToCenter
def getRFCenters(sp, params, type='connected'):
numColumns = np.product(sp.getColumnDimensions())
dimensions = (params['nX'], params['nY'])
meanCoordinates = np.zeros((numColumns, 2))
avgDistToCenter = np.zeros((numColumns, 2))
for columnIndex in range(numColumns):
receptiveField = np.zeros((sp.getNumInputs(), ))
if type == 'connected':
sp.getConnectedSynapses(columnIndex, receptiveField)
elif type == 'potential':
sp.getPotential(columnIndex, receptiveField)
else:
raise RuntimeError('unknown RF type')
connectedSynapseIndex = np.where(receptiveField)[0]
if len(connectedSynapseIndex) == 0:
continue
coordinates = []
for synapseIndex in connectedSynapseIndex:
coordinate = coordinatesFromIndex(synapseIndex, dimensions)
coordinates.append(coordinate)
coordinates = np.array(coordinates)
coordinates = coordinates.astype('float32')
angularCoordinates = np.array(coordinates)
angularCoordinates[:, 0] = coordinates[:, 0] / params['nX'] * 2 * np.pi
angularCoordinates[:, 1] = coordinates[:, 1] / params['nY'] * 2 * np.pi
for i in range(2):
meanCoordinate = np.arctan2(
np.sum(np.sin(angularCoordinates[:, i])),
np.sum(np.cos(angularCoordinates[:, i])))
if meanCoordinate < 0:
meanCoordinate += 2 * np.pi
dist2Mean = angularCoordinates[:, i] - meanCoordinate
dist2Mean = np.arctan2(np.sin(dist2Mean), np.cos(dist2Mean))
dist2Mean = np.max(np.abs(dist2Mean))
meanCoordinate *= dimensions[i] / (2 * np.pi)
dist2Mean *= dimensions[i] / (2 * np.pi)
avgDistToCenter[columnIndex, i] = dist2Mean
meanCoordinates[columnIndex, i] = meanCoordinate
return meanCoordinates, avgDistToCenter
def testCoordinatesFromIndex(self):
self.assertEqual([0], coordinatesFromIndex(0, [100]));
self.assertEqual([50], coordinatesFromIndex(50, [100]));
self.assertEqual([99], coordinatesFromIndex(99, [100]));
self.assertEqual([0, 0], coordinatesFromIndex(0, [100, 80]));
self.assertEqual([0, 10], coordinatesFromIndex(10, [100, 80]));
self.assertEqual([1, 0], coordinatesFromIndex(80, [100, 80]));
self.assertEqual([1, 10], coordinatesFromIndex(90, [100, 80]));
self.assertEqual([0, 0, 0], coordinatesFromIndex(0, [100, 10, 8]));
self.assertEqual([0, 0, 7], coordinatesFromIndex(7, [100, 10, 8]));
self.assertEqual([0, 1, 0], coordinatesFromIndex(8, [100, 10, 8]));
self.assertEqual([1, 0, 0], coordinatesFromIndex(80, [100, 10, 8]));
self.assertEqual([1, 1, 0], coordinatesFromIndex(88, [100, 10, 8]));
self.assertEqual([1, 1, 1], coordinatesFromIndex(89, [100, 10, 8]));
def testCoordinatesFromIndex(self):
self.assertEquals([0], coordinatesFromIndex(0, [100]))
self.assertEquals([50], coordinatesFromIndex(50, [100]))
self.assertEquals([99], coordinatesFromIndex(99, [100]))
self.assertEquals([0, 0], coordinatesFromIndex(0, [100, 80]))
self.assertEquals([0, 10], coordinatesFromIndex(10, [100, 80]))
self.assertEquals([1, 0], coordinatesFromIndex(80, [100, 80]))
self.assertEquals([1, 10], coordinatesFromIndex(90, [100, 80]))
self.assertEquals([0, 0, 0], coordinatesFromIndex(0, [100, 10, 8]))
self.assertEquals([0, 0, 7], coordinatesFromIndex(7, [100, 10, 8]))
self.assertEquals([0, 1, 0], coordinatesFromIndex(8, [100, 10, 8]))
self.assertEquals([1, 0, 0], coordinatesFromIndex(80, [100, 10, 8]))
self.assertEquals([1, 1, 0], coordinatesFromIndex(88, [100, 10, 8]))
self.assertEquals([1, 1, 1], coordinatesFromIndex(89, [100, 10, 8]))