def testRandomSPDoesNotLearn(self):
sp = FlatSpatialPooler(inputShape=5,
coincidencesShape=10,
randomSP=True)
inputArray = (numpy.random.rand(5) > 0.5).astype(uintType)
activeArray = numpy.zeros(sp._numColumns).astype(realType)
# Should start off at 0
self.assertEqual(sp._iterationNum, 0)
self.assertEqual(sp._iterationLearnNum, 0)
# Store the initialized state
initialPerms = copy(sp._permanences)
sp.compute(inputArray, False, activeArray)
# Should have incremented general counter but not learning counter
self.assertEqual(sp._iterationNum, 1)
self.assertEqual(sp._iterationLearnNum, 0)
# Should not learn even if learning set to True
sp.compute(inputArray, True, activeArray)
self.assertEqual(sp._iterationNum, 2)
self.assertEqual(sp._iterationLearnNum, 0)
# Check the initial perm state was not modified either
self.assertEqual(sp._permanences, initialPerms)
def setUp(self):
self._sp = FlatSpatialPooler(
numInputs=5,
numColumns=5,
localAreaDensity=0.1,
numActiveColumnsPerInhArea=-1,
stimulusThreshold=1,
minDistance=0.0,
seed=-1,
spVerbosity=0,
)
def testRandomSPDoesNotLearn(self):
sp = FlatSpatialPooler(inputShape=5,
coincidencesShape=10,
randomSP=True)
inputArray = (numpy.random.rand(5) > 0.5).astype(uintType)
activeArray = numpy.zeros(sp._numColumns).astype(realType)
# Should start off at 0
self.assertEqual(sp._iterationNum, 0)
self.assertEqual(sp._iterationLearnNum, 0)
# Store the initialized state
initialPerms = copy(sp._permanences)
sp.compute(inputArray, False, activeArray)
# Should have incremented general counter but not learning counter
self.assertEqual(sp._iterationNum, 1)
self.assertEqual(sp._iterationLearnNum, 0)
# Should not learn even if learning set to True
sp.compute(inputArray, True, activeArray)
self.assertEqual(sp._iterationNum, 2)
self.assertEqual(sp._iterationLearnNum, 0)
# Check the initial perm state was not modified either
self.assertEqual(sp._permanences, initialPerms)
def setUp(self):
self._sp = FlatSpatialPooler(
inputShape=(5,1),
coincidencesShape=(10,1))
def testActiveColumnsEqualNumActive(self):
'''
After feeding in a record the number of active columns should
always be equal to numActivePerInhArea
'''
for i in [1, 10, 50]:
numActive = i
inputShape = 10
sp = FlatSpatialPooler(inputShape=inputShape,
coincidencesShape=100,
numActivePerInhArea=numActive)
inputArray = (numpy.random.rand(inputShape) > 0.5).astype(uintType)
inputArray2 = (numpy.random.rand(inputShape) > 0.8).astype(uintType)
activeArray = numpy.zeros(sp._numColumns).astype(realType)
# Random SP
sp._randomSP = True
sp.compute(inputArray, False, activeArray)
sp.compute(inputArray2, False, activeArray)
self.assertEqual(sum(activeArray), numActive)
# Default, learning on
sp._randomSP = False
sp.compute(inputArray, True, activeArray)
sp.compute(inputArray2, True, activeArray)
self.assertEqual(sum(activeArray), numActive)
def testActiveColumnsEqualNumActive(self):
'''
After feeding in a record the number of active columns should
always be equal to numActivePerInhArea
'''
for i in [1, 10, 50]:
numActive = i
inputShape = 10
sp = FlatSpatialPooler(inputShape=inputShape,
coincidencesShape=100,
numActivePerInhArea=numActive)
inputArray = (numpy.random.rand(inputShape) > 0.5).astype(uintType)
inputArray2 = (numpy.random.rand(inputShape) > 0.8).astype(uintType)
activeArray = numpy.zeros(sp._numColumns).astype(realType)
# Random SP
sp._randomSP = True
sp.compute(inputArray, False, activeArray)
sp.compute(inputArray2, False, activeArray)
self.assertEqual(sum(activeArray), numActive)
# Default, learning on
sp._randomSP = False
sp.compute(inputArray, True, activeArray)
sp.compute(inputArray2, True, activeArray)
self.assertEqual(sum(activeArray), numActive)