def initMixtureModel(data, hyperParams):
# Initialize parameter data structs
C = hyperParams.C
K = hyperParams.K
mixture = [1.0 / C] * C
multinomials = []
for c in range(0, C):
randomMultinomial = ST.drawFromDirichlet(hyperParams.componentDirich)
multinomials.append(randomMultinomial)
return MME.MultinomialMixtureModel(C, K, multinomials, mixture)
def readSingleTreeFromInfile(reader):
next = reader.next()
mixture = map(float, next)
if (len(mixture) == 0): return None # We've reached a terminal node
multinomials = []
for i in range(0, len(mixture)):
row = reader.next()
multinomials.append(map(float, row))
K = 2
if (len(multinomials) > 0): K = len(multinomials[0])
return MME.MultinomialMixtureModel(len(mixture), K, multinomials, mixture)
def initMixtureModel(data, hyperParams):
# Initialize parameter data structs
C = hyperParams.C
K = hyperParams.K
mixture = [1.0 / C] * C
multinomials = []
for c in range(0, C):
multinomials.append([0.0] * K)
denominator = sum(data[c]) + sum(hyperParams.componentDirich)
for k in range(0, K):
numerator = float(data[c][k]) + hyperParams.componentDirich[k]
multinomials[c][k] = numerator / denominator
return MME.MultinomialMixtureModel(C, K, multinomials, mixture)
def updateMixtureModel(data, params, hyperParams, batchSize, learnRate):
minibatch = makeMinibatch(data, batchSize)
# Initialize parameter data structs
C = params.C
K = params.K
mixtureCounts = [0.] * C
multinomialCounts = []
for c in range(0, C):
mixtureCounts[c] += hyperParams.mixtureDirich[c]
multinomialCounts.append([0.] * K)
for k in range(0, K):
multinomialCounts[c][k] += hyperParams.componentDirich[k]
# Loop through the data and update param data structs
for row in minibatch:
cProbs = MME.getComponentProbabilitiesForCounts(row, params)
for c in range(0, C):
mixtureCounts[c] += cProbs[c]
for k in range(0, K):
multinomialCounts[c][k] += row[k] * cProbs[c]
# Update the mixture probabilities
sumMixtureCounts = sum(mixtureCounts)
for c in range(0, C):
mixtureCounts[c] /= sumMixtureCounts
mixtureCounts[c] = learnRate * mixtureCounts[c] + (
1 - learnRate) * params.mixture[c]
# Update the multinomial probabilities
for c in range(0, C):
multinomialSum = sum(multinomialCounts[c])
for k in range(0, K):
multinomialCounts[c][k] /= multinomialSum
multinomialCounts[c][k] = learnRate * multinomialCounts[c][k] + (
1 - learnRate) * params.multinomials[c][k]
return MME.MultinomialMixtureModel(C, K, multinomialCounts, mixtureCounts)