def buildMixtureTreeMaxKL(data, K, iterations, maxKL, branchesPerNode = 2, multinomial = None):
logging.info("Dataset Size: " + str(len(data)))
if (multinomial):
(worstKL, worstN) = MME.worstFitForSingleMultinomial(data, multinomial)
logging.info("Worst KL found in branch: " + str(worstKL))
if (worstKL < maxKL): return None
if (len(data) < branchesPerNode): return None
# hyperparameters are fixed here:
hyperP = MME.MultinomialMixtureModelHyperparams(branchesPerNode, K, [1.0 / branchesPerNode]*branchesPerNode, [1.0 / K]*K)
mixtureModel = MME.computeDirichletMixture(data, hyperP, iterations)
smallerDatasets = []
for c in range(0, hyperP.C): smallerDatasets.append([])
for counts in data:
c = MME.assignComponentToCounts(counts, mixtureModel)
smallerDatasets[c].append(counts)
treeModel = MultinomialMixtureTree(mixtureModel)
for c in range(0, hyperP.C):
smallerDataset = smallerDatasets[c]
child = buildMixtureTreeMaxKL(smallerDataset, K, iterations, maxKL, branchesPerNode, mixtureModel.multinomials[c])
treeModel.mixtureNodes[c] = child
return treeModel
def buildSimpleMixtureTree(data, K, iterations, height, branchesPerNode = 2):
if (height == 0): return None
# hyperparameters are fixed here:
hyperP = MME.MultinomialMixtureModelHyperparams(branchesPerNode, K, [1.0 / branchesPerNode]*branchesPerNode, [1.0 / K]*K)
mixtureModel = MME.computeDirichletMixture(data, hyperP, iterations)
smallerDatasets = []
for c in range(0, hyperP.C): smallerDatasets.append([])
for counts in data:
c = MME.assignComponentToCounts(counts, mixtureModel)
smallerDatasets[c].append(counts)
treeModel = MultinomialMixtureTree(mixtureModel)
for c in range(0, hyperP.C):
smallerDataset = smallerDatasets[c]
child = buildSimpleMixtureTree(smallerDataset, K, iterations, height - 1, branchesPerNode)
treeModel.mixtureNodes[c] = child
return treeModel
dataset.append(map(int, splitrow))
#for n in range(0, len(dataset)):
# counts = dataset[n]
# print str(n) + "\t" + str(MME.assignComponentToCounts(counts, model))
# print file for google docs
#print "component\t",
#for i in range(0, C): print str(i) + "\t",
#print ""
#print "prior\t" + "\t".join(map(str, finalModel.mixture))
#
#for k in range(0, 168):
# print str(k) + "\t",
# for i in range(0, C):
# print str(finalModel.multinomials[i][k]) + "\t",
# print ""
rowInfoList = []
N = 0
for row in dataset:
c = MME.assignComponentToCounts(row, model)
klDiv = MME.klTest(row, model, c)
rowInfoList.append([N, c, klDiv])
N += 1
sortedRows = sorted(rowInfoList, key=(lambda x: -x[2]))
print "row\tmodel\tklDivergence"
for row in sortedRows:
print "\t".join(map(str, row))
for row in sys.stdin:
splitrow = row.split("\t")
dataset.append(map(int, splitrow))
#for n in range(0, len(dataset)):
# counts = dataset[n]
# print str(n) + "\t" + str(MME.assignComponentToCounts(counts, model))
# print file for google docs
#print "component\t",
#for i in range(0, C): print str(i) + "\t",
#print ""
#print "prior\t" + "\t".join(map(str, finalModel.mixture))
#
#for k in range(0, 168):
# print str(k) + "\t",
# for i in range(0, C):
# print str(finalModel.multinomials[i][k]) + "\t",
# print ""
rowInfoList = []
N = 0
for row in dataset:
c = MME.assignComponentToCounts(row, model)
klDiv = MME.klTest(row, model.multinomials[c])
rowInfoList.append([N, c, klDiv, sum(row)])
N += 1
print "row\tmodel\tklDivergence\tNumber of Data Points"
for row in rowInfoList:
print "\t".join(map(str, row))