def initialize(self, data=[], noise=1):
""" Initialize the model. It construct the baseModel by assembling all
quantities from the components.
Args:
noise: the initial guess of the variance of the observation noise.
"""
if len(self.staticComponents) == 0 and \
len(self.dynamicComponents) == 0 and \
len(self.automaticComponents) == 0:
raise NameError('The model must contain at least' +
' one component')
# construct transition, evaluation, prior state, prior covariance
if self._printInfo:
print('Initializing models...')
transition = None
evaluation = None
state = None
sysVar = None
self.discount = np.array([])
# first construct for the static components
# the evaluation will be treated separately for static or dynamic
# as the latter one will change over time
currentIndex = 0 # used for compute the index
for i in self.staticComponents:
comp = self.staticComponents[i]
transition = mt.matrixAddInDiag(transition, comp.transition)
evaluation = mt.matrixAddByCol(evaluation,
comp.evaluation)
state = mt.matrixAddByRow(state, comp.meanPrior)
sysVar = mt.matrixAddInDiag(sysVar, comp.covPrior)
self.discount = np.concatenate((self.discount, comp.discount))
self.componentIndex[i] = (currentIndex, currentIndex + comp.d - 1)
currentIndex += comp.d
# if the model contains the dynamic part, we add the dynamic components
if len(self.dynamicComponents) > 0:
self.dynamicEvaluation = None
for i in self.dynamicComponents:
comp = self.dynamicComponents[i]
comp.updateEvaluation(0)
transition = mt.matrixAddInDiag(transition, comp.transition)
evaluation = mt.matrixAddByCol(evaluation,
comp.evaluation)
state = mt.matrixAddByRow(state, comp.meanPrior)
sysVar = mt.matrixAddInDiag(sysVar, comp.covPrior)
self.discount = np.concatenate((self.discount, comp.discount))
self.componentIndex[i] = (currentIndex,
currentIndex + comp.d - 1)
currentIndex += comp.d
# if the model contains the automatic dynamic part, we add
# them to the builder
if len(self.automaticComponents) > 0:
self.automaticEvaluation = None
for i in self.automaticComponents:
comp = self.automaticComponents[i]
comp.updateEvaluation(0, data)
transition = mt.matrixAddInDiag(transition, comp.transition)
evaluation = mt.matrixAddByCol(evaluation,
comp.evaluation)
state = mt.matrixAddByRow(state, comp.meanPrior)
sysVar = mt.matrixAddInDiag(sysVar, comp.covPrior)
self.discount = np.concatenate((self.discount, comp.discount))
self.componentIndex[i] = (currentIndex,
currentIndex + comp.d - 1)
currentIndex += comp.d
self.statePrior = state
self.sysVarPrior = sysVar
self.noiseVar = np.matrix(noise)
self.model = baseModel(transition=transition,
evaluation=evaluation,
noiseVar=np.matrix(noise),
sysVar=sysVar,
state=state,
df=self.initialDegreeFreedom)
self.model.initializeObservation()
# compute the renew period
if self.renewDiscount is None:
self.renewDiscount = np.min(self.discount)
if self.renewDiscount < 1.0 - 1e-8:
self.renewTerm = np.log(0.001 * (1 - self.renewDiscount)) \
/ np.log(self.renewDiscount)
self.initialized = True
if self._printInfo:
print('Initialization finished.')
def initialize(self, data=[], noise=1):
""" Initialize the model. It construct the baseModel by assembling all
quantities from the components.
Args:
noise: the initial guess of the variance of the observation noise.
"""
if len(self.staticComponents) == 0 and \
len(self.dynamicComponents) == 0 and \
len(self.automaticComponents) == 0:
raise NameError('The model must contain at least' +
' one component')
# construct transition, evaluation, prior state, prior covariance
if self._printInfo:
print('Initializing models...')
transition = None
evaluation = None
state = None
sysVar = None
self.discount = np.array([])
# first construct for the static components
# the evaluation will be treated separately for static or dynamic
# as the latter one will change over time
currentIndex = 0 # used for compute the index
for i in self.staticComponents:
comp = self.staticComponents[i]
transition = mt.matrixAddInDiag(transition, comp.transition)
evaluation = mt.matrixAddByCol(evaluation, comp.evaluation)
state = mt.matrixAddByRow(state, comp.meanPrior)
sysVar = mt.matrixAddInDiag(sysVar, comp.covPrior)
self.discount = np.concatenate((self.discount, comp.discount))
self.componentIndex[i] = (currentIndex, currentIndex + comp.d - 1)
currentIndex += comp.d
# if the model contains the dynamic part, we add the dynamic components
if len(self.dynamicComponents) > 0:
self.dynamicEvaluation = None
for i in self.dynamicComponents:
comp = self.dynamicComponents[i]
comp.updateEvaluation(0)
transition = mt.matrixAddInDiag(transition, comp.transition)
evaluation = mt.matrixAddByCol(evaluation, comp.evaluation)
state = mt.matrixAddByRow(state, comp.meanPrior)
sysVar = mt.matrixAddInDiag(sysVar, comp.covPrior)
self.discount = np.concatenate((self.discount, comp.discount))
self.componentIndex[i] = (currentIndex,
currentIndex + comp.d - 1)
currentIndex += comp.d
# if the model contains the automatic dynamic part, we add
# them to the builder
if len(self.automaticComponents) > 0:
self.automaticEvaluation = None
for i in self.automaticComponents:
comp = self.automaticComponents[i]
comp.updateEvaluation(0, data)
transition = mt.matrixAddInDiag(transition, comp.transition)
evaluation = mt.matrixAddByCol(evaluation, comp.evaluation)
state = mt.matrixAddByRow(state, comp.meanPrior)
sysVar = mt.matrixAddInDiag(sysVar, comp.covPrior)
self.discount = np.concatenate((self.discount, comp.discount))
self.componentIndex[i] = (currentIndex,
currentIndex + comp.d - 1)
currentIndex += comp.d
self.statePrior = state
self.sysVarPrior = sysVar
self.noiseVar = np.matrix(noise)
self.model = baseModel(transition=transition,
evaluation=evaluation,
noiseVar=np.matrix(noise),
sysVar=sysVar,
state=state,
df=self.initialDegreeFreedom)
self.model.initializeObservation()
# compute the renew period
if self.renewDiscount is None:
self.renewDiscount = np.min(self.discount)
if self.renewDiscount < 1.0 - 1e-8:
self.renewTerm = np.log(0.001 * (1 - self.renewDiscount)) \
/ np.log(self.renewDiscount)
self.initialized = True
if self._printInfo:
print('Initialization finished.')