def setUp(self):
self.unionTemporalPooler = UnionTemporalPooler(inputDimensions=5,
columnDimensions=5,
potentialRadius=16,
potentialPct=0.9,
globalInhibition=True,
localAreaDensity=-1.0,
numActiveColumnsPerInhArea=2.0,
stimulusThreshold=2,
synPermInactiveDec=0.01,
synPermActiveInc=0.03,
synPermConnected=0.3,
minPctOverlapDutyCycle=0.001,
minPctActiveDutyCycle=0.001,
dutyCyclePeriod=1000,
maxBoost=1.0,
seed=42,
spVerbosity=0,
wrapAround=True,
# union_temporal_pooler.py parameters
activeOverlapWeight=1.0,
predictedActiveOverlapWeight=10.0,
maxUnionActivity=0.20,
exciteFunctionType='Fixed',
decayFunctionType='NoDecay')
class UnionTemporalPoolerTest(unittest.TestCase):
def setUp(self):
self.unionTemporalPooler = UnionTemporalPooler(inputDimensions=5,
columnDimensions=5,
potentialRadius=16,
potentialPct=0.9,
globalInhibition=True,
localAreaDensity=-1.0,
numActiveColumnsPerInhArea=2.0,
stimulusThreshold=2,
synPermInactiveDec=0.01,
synPermActiveInc=0.03,
synPermConnected=0.3,
minPctOverlapDutyCycle=0.001,
minPctActiveDutyCycle=0.001,
dutyCyclePeriod=1000,
maxBoost=1.0,
seed=42,
spVerbosity=0,
wrapAround=True,
# union_temporal_pooler.py parameters
activeOverlapWeight=1.0,
predictedActiveOverlapWeight=10.0,
maxUnionActivity=0.20,
exciteFunctionType='Fixed',
decayFunctionType='NoDecay')
def testDecayPoolingActivationDefaultDecayRate(self):
self.unionTemporalPooler._poolingActivation = numpy.array([0, 1, 2, 3, 4],
dtype=REAL_DTYPE)
expected = numpy.array([0, 1, 2, 3, 4], dtype=REAL_DTYPE)
result = self.unionTemporalPooler._decayPoolingActivation()
print result
self.assertTrue(numpy.array_equal(expected, result))
def testAddToPoolingActivation(self):
activeCells = numpy.array([1, 3, 4])
overlaps = numpy.array([0.123, 0.0, 0.0, 0.456, 0.789])
expected = [0.0, 10.0, 0.0, 10.0, 10.0]
result = self.unionTemporalPooler._addToPoolingActivation(activeCells, overlaps)
self.assertTrue(numpy.allclose(expected, result))
def testAddToPoolingActivationExistingActivation(self):
self.unionTemporalPooler._poolingActivation = numpy.array([0, 1, 2, 3, 4],
dtype=REAL_DTYPE)
activeCells = numpy.array([1, 3, 4])
# [ 0, 1, 0, 1, 1]
overlaps = numpy.array([0.123, 0.0, 0.0, 0.456, 0.789])
expected = [0.0, 11.0, 2.0, 13, 14]
result = self.unionTemporalPooler._addToPoolingActivation(activeCells, overlaps)
self.assertTrue(numpy.allclose(expected, result))
def testGetMostActiveCellsUnionSizeZero(self):
self.unionTemporalPooler._poolingActivation = numpy.array([0, 1, 2, 3, 4],
dtype=REAL_DTYPE)
self.unionTemporalPooler._maxUnionCells = 0
result = self.unionTemporalPooler._getMostActiveCells()
self.assertEquals(len(result), 0)
def testGetMostActiveCellsRegular(self):
self.unionTemporalPooler._poolingActivation = numpy.array([0, 1, 2, 3, 4],
dtype=REAL_DTYPE)
result = self.unionTemporalPooler._getMostActiveCells()
self.assertEquals(len(result), 1)
self.assertEquals(result[0], 4)
def testGetMostActiveCellsIgnoreZeros(self):
self.unionTemporalPooler._poolingActivation = numpy.array([0, 0, 0, 3, 4],
dtype=REAL_DTYPE)
self.unionTemporalPooler._maxUnionCells = 3
result = self.unionTemporalPooler._getMostActiveCells()
self.assertEquals(len(result), 2)
self.assertEquals(result[0], 3)
self.assertEquals(result[1], 4)