def memo_factorial_term_for_merge(self, LP, mPairIDs):
if mPairIDs is None:
logFactZMerge = NumericHardUtil.colwisesumLogFactorial_allpairs(
LP['hard_asgn'])
else:
logFactZMerge = NumericHardUtil.colwisesumLogFactorial_specificpairs(
LP['hard_asgn'], mPairIDs)
return logFactZMerge
def get_hard_word_variational(self, Data, LP):
''' Update and return word-topic assignment variational parameters
'''
LP['word_variational'] = NumericHardUtil.toHardAssignmentMatrix(
LP['word_variational'])
return LP
"""
def calc_local_params(self, Data, LP, **kwargs):
""" Calculate local parameters for each data item and each component.
This is part of the E-step.
Args
-------
Data : bnpy data object with Data.nObs observations
LP : local param dict with fields
E_log_soft_ev : Data.nObs x K array
E_log_soft_ev[n,k] = log p(data obs n | comp k)
Returns
-------
LP : local param dict with fields
resp : Data.nObs x K array whose rows sum to one
resp[n,k] = posterior responsibility that comp. k has for data n
"""
lpr = LP["E_log_soft_ev"]
lpr += self.Elogw
LP["resp"] = NumericHardUtil.toHardAssignmentMatrix(lpr)
assert np.allclose(LP["resp"].sum(axis=1), 1)
return LP
def calc_local_params(self, Data, LP,
nCoordAscentItersLP=20, convThrLP=0.01,
doOnlySomeDocsLP=True, **kwargs):
''' Calculate document-specific quantities (E-step) using hard assignments.
Alternate updates to two terms until convergence
(1) Approx posterior on topic-token assignment
q(word_variational | word_token_variables)
(2) Approx posterior on doc-topic probabilities
q(doc_variational | document_topic_variables)
Returns
-------
LP : local params dict
'''
# First, run soft assignments for nCoordAscentIters
LP = self.calc_local_params_fast(Data, LP,
nCoordAscentItersLP,
convThrLP,
doOnlySomeDocsLP,
)
# Next, find hard assignments
LP['hard_asgn'] = NumericHardUtil.findMode_Mult(
Data.word_count,
LP['word_variational']
)
# Update DocTopicCount to use hard assignments
for d in xrange(Data.nDoc):
start = Data.doc_range[d,0]
stop = Data.doc_range[d,1]
LP['DocTopicCount'][d,:] = np.sum(LP['hard_asgn'][start:stop], axis=0)
# Update alphPi to use hard assignments
LP = self.get_doc_variational(Data, LP)
LP = self.calc_ElogPi(LP)
return LP
def get_hard_word_variational(self, Data, LP):
""" Update and return word-topic assignment variational parameters
"""
LP["word_variational"] = NumericHardUtil.toHardAssignmentMatrix(LP["word_variational"])
return LP
"""
def E_logfactorialZ(self, Data, LP): logFactData = NumericHardUtil.colwisesumLogFactorial(Data.word_count) logFactHardAsgn = NumericHardUtil.colwisesumLogFactorial(LP['hard_asgn']) return logFactData, logFactHardAsgn
