def setPostFactors(self,
obsModel=None,
SS=None,
LP=None,
Data=None,
**param_kwargs):
''' Set attribute Post to provided values.
'''
self.ClearCache()
if obsModel is not None:
if hasattr(obsModel, 'Post'):
self.Post = obsModel.Post.copy()
self.K = self.Post.K
else:
self.setPostFromEstParams(obsModel.EstParams)
return
if LP is not None and Data is not None:
SS = self.calcSummaryStats(Data, None, LP)
if SS is not None:
self.updatePost(SS)
else:
self.Post = DiagGaussX.packParamBagForPost(**param_kwargs)
self.Post = RegressY.packParamBagForPost(Post=self.Post,
**param_kwargs)
self.K = self.Post.K
def calcELBO_Memoized(self, SS, returnVec=0, afterMStep=False, **kwargs):
""" Calculate obsModel's objective using suff stats SS and Post.
Args
-------
SS : bnpy SuffStatBag
afterMStep : boolean flag
if 1, elbo calculated assuming M-step just completed
Returns
-------
elbo_K : scalar float
Equal to E[ log p(x) + log p(phi) - log q(phi)]
"""
elbo_XModel = DiagGaussX.calcELBOFromSSAndPost(SS=SS,
Post=self.Post,
Prior=self.Prior,
returnVec=returnVec,
afterMStep=afterMStep)
elbo_YModel = RegressY.calcELBOFromSSAndPost(SS=SS,
Post=self.Post,
Prior=self.Prior,
returnVec=returnVec,
afterMStep=afterMStep)
return elbo_XModel + elbo_YModel
def calcSummaryStats(self, Data, SS, LP, **kwargs):
''' Calculate summary statistics for given dataset and local parameters
Returns
--------
SS : SuffStatBag object, with K components.
'''
SS = DiagGaussX.calcSummaryStats(Data, SS, LP, **kwargs)
SS = RegressY.calcSummaryStats(Data, SS, LP, **kwargs)
return SS
def calcLogSoftEvMatrix_FromPost(self, Data, **kwargs):
''' Compute expected log soft evidence of each item under each cluster
Returns
-------
E_log_soft_ev_NK : 2D array, size N x K
'''
E_log_soft_ev_NK = DiagGaussX.calcLogSoftEvMatrix_FromPost(
Data, Post=self.Post)
if hasattr(Data, 'Y'):
# Only incorporate Y attribute if present
E_log_soft_ev_NK = RegressY.calcLogSoftEvMatrix_FromPost(
Data, Post=self.Post, E_log_soft_ev_NK=E_log_soft_ev_NK)
return E_log_soft_ev_NK
def calcHardMergeGap(self, SS, kA, kB):
''' Calculate change in ELBO after a hard merge applied to this model
Returns
---------
gap : scalar real, indicates change in ELBO after merge of kA, kB
'''
gapX, _, _ = DiagGaussX.calcHardMergeGapForPair(SS=SS,
Post=self.Post,
Prior=self.Prior,
kA=kA,
kB=kB)
gapY, _, _ = RegressY.calcHardMergeGapForPair(SS=SS,
Post=self.Post,
Prior=self.Prior,
kA=kA,
kB=kB)
return gapX + gapY
def updatePost(self, SS):
''' Update attribute Post for all comps given suff stats.
Optimizes the variational objective for approximating the posterior
Post Condition
--------------
Attributes K and Post updated in-place.
'''
self.ClearCache()
if not hasattr(self, 'Post') or SS.K != self.Post.K:
self.Post = None
self.Post = DiagGaussX.calcPostParamsFromSS(SS=SS,
Prior=self.Prior,
Post=self.Post)
self.Post = RegressY.calcPostParamsFromSS(SS=SS,
Prior=self.Prior,
Post=self.Post)
self.K = SS.K
def __init__(self, inferType='VB', D=0, Data=None, **PriorArgs):
''' Initialize bare obsmodel with valid prior hyperparameters.
Returns
-------
obsmodel : bare observation model
Resulting object lacks either EstParams or Post attributes.
which must be created separately (see init_global_params).
'''
if Data is not None:
self.D = Data.dim
else:
self.D = int(D)
self.E = self.D + 1
self.K = 0
self.inferType = inferType
self.Cache = dict()
PriorArgs['D'] = self.D
PriorArgs['E'] = self.E
self.Prior = DiagGaussX.createParamBagForPrior(Data, **PriorArgs)
self.Prior = RegressY.createParamBagForPrior(Data,
Prior=self.Prior,
**PriorArgs)
def get_info_string_prior(self):
return DiagGaussX.getStringSummaryOfPrior(self.Prior) \
+ RegressY.getStringSummaryOfPrior(self.Prior)