def vizInterSequenceClusters(traces, outputDir, cellsType, numCells,
numClasses, ignoreNoise=True):
clusters = assignClusters(traces)
# compare inter-cluster distance over time
numRptsPerCategory = {}
categories = np.unique(traces['actualCategory'])
repetition = np.array(clusters['repetition'])
for category in categories:
numRptsPerCategory[category] = np.max(
repetition[np.array(traces['actualCategory']) == category])
SDRclusters = []
clusterAssignments = []
numRptsMin = np.min(numRptsPerCategory.values()).astype('int32')
for rpt in range(numRptsMin + 1):
idx0 = np.logical_and(np.array(traces['actualCategory']) == 0,
repetition == rpt)
idx1 = np.logical_and(np.array(traces['actualCategory']) == 1,
repetition == rpt)
idx2 = np.logical_and(np.array(traces['actualCategory']) == 2,
repetition == rpt)
c0slice = [traces['tmPredictedActiveCells'][i] for i in range(len(idx0)) if
idx0[i]]
c1slice = [traces['tmPredictedActiveCells'][i] for i in range(len(idx1)) if
idx1[i]]
c2slice = [traces['tmPredictedActiveCells'][i] for i in range(len(idx2)) if
idx2[i]]
if not ignoreNoise:
SDRclusters.append(c0slice)
clusterAssignments.append(0)
SDRclusters.append(c1slice)
clusterAssignments.append(1)
SDRclusters.append(c2slice)
clusterAssignments.append(2)
print " Presentation #{}: ".format(rpt)
if not ignoreNoise:
d01 = clusterDist(c0slice, c1slice, numCells)
print '=> d(c0, c1): %s' % d01
d02 = clusterDist(c0slice, c2slice, numCells)
print '=> d(c0, c2): %s' % d02
d12 = clusterDist(c1slice, c2slice, numCells)
print '=> d(c1, c2): %s' % d12
npos, distanceMat = projectClusters2D(SDRclusters, numCells)
title = 'Inter-sequence clusters in 2D (using %s)' % cellsType
outputFile = '%s/%s' % (outputDir, title)
viz2DProjection(title, outputFile, numClasses, clusterAssignments, npos)
title = 'Inter-sequence clusters distances (using %s)' % cellsType
outputFile = '%s/%s' % (outputDir, title)
plotDistanceMat(distanceMat, title, outputFile)
def computeClusterDistanceMat(sdrClusters):
"""
Compute distance matrix between clusters of SDRs
:param sdrClusters: list of sdr clusters,
each cluster is a list of SDRs
each SDR is a list of active indices
:return: distance matrix
"""
numClusters = len(sdrClusters)
distanceMat = np.zeros((numClusters, numClusters), dtype=np.float64)
for i in range(numClusters):
for j in range(i, numClusters):
distanceMat[i, j] = clusterDist(sdrClusters[i], sdrClusters[j])
distanceMat[j, i] = distanceMat[i, j]
return distanceMat
def computeClusterDistanceMat(sdrClusters, numCells):
"""
Compute distance matrix between clusters of SDRs
:param sdrClusters: list of sdr clusters,
each cluster is a list of SDRs
each SDR is a list of active indices
:return: distance matrix
"""
numClusters = len(sdrClusters)
distanceMat = np.zeros((numClusters, numClusters), dtype=np.float64)
for i in range(numClusters):
for j in range(i, numClusters):
distanceMat[i, j] = clusterDist(sdrClusters[i], sdrClusters[j],
numCells)
distanceMat[j, i] = distanceMat[i, j]
return distanceMat
def vizInterSequenceClusters(traces,
outputDir,
cellsType,
numCells,
numClasses,
ignoreNoise=True):
clusters = assignClusters(traces)
# compare inter-cluster distance over time
numRptsPerCategory = {}
categories = np.unique(traces['actualCategory'])
repetition = np.array(clusters['repetition'])
for category in categories:
numRptsPerCategory[category] = np.max(
repetition[np.array(traces['actualCategory']) == category])
SDRclusters = []
clusterAssignments = []
numRptsMin = np.min(numRptsPerCategory.values()).astype('int32')
for rpt in range(numRptsMin + 1):
idx0 = np.logical_and(
np.array(traces['actualCategory']) == 0, repetition == rpt)
idx1 = np.logical_and(
np.array(traces['actualCategory']) == 1, repetition == rpt)
idx2 = np.logical_and(
np.array(traces['actualCategory']) == 2, repetition == rpt)
c0slice = [
traces['tmPredictedActiveCells'][i] for i in range(len(idx0))
if idx0[i]
]
c1slice = [
traces['tmPredictedActiveCells'][i] for i in range(len(idx1))
if idx1[i]
]
c2slice = [
traces['tmPredictedActiveCells'][i] for i in range(len(idx2))
if idx2[i]
]
if not ignoreNoise:
SDRclusters.append(c0slice)
clusterAssignments.append(0)
SDRclusters.append(c1slice)
clusterAssignments.append(1)
SDRclusters.append(c2slice)
clusterAssignments.append(2)
print " Presentation #{}: ".format(rpt)
if not ignoreNoise:
d01 = clusterDist(c0slice, c1slice, numCells)
print '=> d(c0, c1): %s' % d01
d02 = clusterDist(c0slice, c2slice, numCells)
print '=> d(c0, c2): %s' % d02
d12 = clusterDist(c1slice, c2slice, numCells)
print '=> d(c1, c2): %s' % d12
npos, distanceMat = projectClusters2D(SDRclusters, numCells)
title = 'Inter-sequence clusters in 2D (using %s)' % cellsType
outputFile = '%s/%s' % (outputDir, title)
viz2DProjection(title, outputFile, numClasses, clusterAssignments, npos)
title = 'Inter-sequence clusters distances (using %s)' % cellsType
outputFile = '%s/%s' % (outputDir, title)
plotDistanceMat(distanceMat, title, outputFile)