def __init__(self,gm,accumulator=None,parameter=InfParam()):
if accumulator is None:
self.accumulator=defaultAccumulator(gm=gm)
else:
self.accumulator=accumulator
kwargs=parameter.kwargs
self.gm_=gm
self.arg_ = None
self.value_ = None
self.initType = kwargs.get('initType', 'localOpt')
def __init__(self, gm, accumulator=None, parameter=InfParam()):
if accumulator is None:
self.accumulator = defaultAccumulator(gm=gm)
else:
self.accumulator = accumulator
kwargs = parameter.kwargs
self.gm_ = gm
self.arg_ = None
self.value_ = None
self.initType = kwargs.get('initType', 'localOpt')
def __init__(self,gm,accumulator=None,parameter=InfParam()):
print "fresh constructor "
if accumulator is None:
self.accumulator=defaultAccumulator(gm=gm)
else:
self.accumulator=accumulator
kwargs=parameter.kwargs
self.gm_=gm
self.solverList = kwargs.get('solvers', [])
self.parameterList = kwargs.get('parameters', [])
self.arg_ = numpy.zeros(gm.numberOfVariables,dtype=numpy.uint64)
def __init__(self, gm, accumulator=None, parameter=InfParam()):
if accumulator is None:
self.accumulator = defaultAccumulator(gm=gm)
else:
self.accumulator = accumulator
kwargs = parameter.kwargs
self.gm_ = gm
self.steps = kwargs.get('steps', 100)
self.fusionSolver = kwargs.get('fuisionSolver', 'lf2')
self.arg_ = None
self.value_ = None
self.fusionMover = inference.adder.minimizer.FusionMover(self.gm_)
self.nLabels = self.gm_.numberOfLabels(0)
self.nVar = self.gm_.numberOfVariables
def __init__(self, gm, accumulator=None, parameter=InfParam()):
if accumulator is None:
self.accumulator = defaultAccumulator(gm=gm)
else:
self.accumulator = accumulator
kwargs = parameter.kwargs
self.gm_ = gm
self.steps = kwargs.get("steps", 100)
self.fusionSolver = kwargs.get("fuisionSolver", "lf2")
self.arg_ = None
self.value_ = None
self.fusionMover = inference.adder.minimizer.FusionMover(self.gm_)
self.nLabels = self.gm_.numberOfLabels(0)
self.nVar = self.gm_.numberOfVariables
def inference_init(self, gm, accumulator=None, parameter=None):
# self._old_init()
# set up basic properties
self.gm = gm
self.operator = gm.operator
if accumulator is None:
self.accumulator = defaultAccumulator(gm)
else:
self.accumulator = accumulator
self._meta_parameter = parameter
# get hyper parameter (as minStCut for graphcut, or the subsolver for
# dualdec.)
hyperParamKeywords = self._infClasses.hyperParameterKeywords
numHyperParams = len(hyperParamKeywords)
userHyperParams = [None] * numHyperParams
collectedHyperParameters = 0
# get the users hyper parameter ( if given)
if (self._meta_parameter is not None):
for hpIndex, hyperParamKeyword in enumerate(hyperParamKeywords):
if hyperParamKeyword in self._meta_parameter.kwargs:
userHyperParams[hpIndex] = self._meta_parameter.kwargs.pop(
hyperParamKeyword)
collectedHyperParameters += 1
# check if ZERO or ALL hyperParamerts have been collected
if collectedHyperParameters != 0 and collectedHyperParameters != numHyperParams:
raise RuntimeError("All or none hyper-parameter must be given")
# check if the WHOLE tuple of hyperParameters is allowed
if collectedHyperParameters != 0:
if tuple(
str(x)
for x in userHyperParams) not in inferenceClasses.implDict:
raise RuntimeError(
"%s is not an allowed hyperParameter\nAllowed hyperParameters are %s"
% (repr(userHyperParams),
repr(inferenceClasses.implDict.keys())))
else:
userHyperParams = defaultHyperParams
try:
# get the selected inference class and the parameter
if (numHyperParams == 0):
self._selectedInfClass, self._selectedInfParamClass = inferenceClasses.implDict[
"__NONE__"][(self.operator, self.accumulator)]
else:
hp = tuple(str(x) for x in userHyperParams)
self._selectedInfClass, self._selectedInfParamClass = inferenceClasses.implDict[
hp][(self.operator, self.accumulator)]
except:
dictStr = str(inferenceClasses.implDict)
raise RuntimeError("given seminring (operator = %s ,accumulator = %s) is not implemented for this solver\n %s" % \
(self.operator, self.accumulator,dictStr))
if self._meta_parameter is None:
self.parameter = self._selectedInfClass._parameter()
self.parameter.set()
else:
self.parameter = to_native_class_converter(
givenValue=self._meta_parameter,
nativeClass=self._selectedInfParamClass)
assert self.parameter is not None
self.inference = self._selectedInfClass(self.gm, self.parameter)
def inference_init(self, gm, accumulator=None, parameter=None):
# self._old_init()
# set up basic properties
self.gm = gm
self.operator = gm.operator
if accumulator is None:
self.accumulator = defaultAccumulator(gm)
else:
self.accumulator = accumulator
self._meta_parameter = parameter
# get hyper parameter (as minStCut for graphcut, or the subsolver for
# dualdec.)
hyperParamKeywords = self._infClasses.hyperParameterKeywords
numHyperParams = len(hyperParamKeywords)
userHyperParams = [None]*numHyperParams
collectedHyperParameters = 0
# get the users hyper parameter ( if given)
if(self._meta_parameter is not None):
for hpIndex, hyperParamKeyword in enumerate(hyperParamKeywords):
if hyperParamKeyword in self._meta_parameter.kwargs:
userHyperParams[hpIndex] = self._meta_parameter.kwargs.pop(
hyperParamKeyword)
collectedHyperParameters += 1
# check if ZERO or ALL hyperParamerts have been collected
if collectedHyperParameters != 0 and collectedHyperParameters != numHyperParams:
raise RuntimeError("All or none hyper-parameter must be given")
# check if the WHOLE tuple of hyperParameters is allowed
if collectedHyperParameters != 0:
if tuple(str(x) for x in userHyperParams) not in inferenceClasses.implDict:
raise RuntimeError("%s is not an allowed hyperParameter\nAllowed hyperParameters are %s" % (
repr(userHyperParams), repr(inferenceClasses.implDict.keys())))
else:
userHyperParams = defaultHyperParams
try:
# get the selected inference class and the parameter
if(numHyperParams == 0):
self._selectedInfClass, self._selectedInfParamClass = inferenceClasses.implDict[
"__NONE__"][(self.operator, self.accumulator)]
else:
hp = tuple(str(x) for x in userHyperParams)
self._selectedInfClass, self._selectedInfParamClass = inferenceClasses.implDict[
hp][(self.operator, self.accumulator)]
except:
dictStr=str(inferenceClasses.implDict)
raise RuntimeError("given seminring (operator = %s ,accumulator = %s) is not implemented for this solver\n %s" % \
(self.operator, self.accumulator,dictStr))
if self._meta_parameter is None:
self.parameter = self._selectedInfClass._parameter()
self.parameter.set()
else:
self.parameter = to_native_class_converter(
givenValue=self._meta_parameter, nativeClass=self._selectedInfParamClass)
assert self.parameter is not None
self.inference = self._selectedInfClass(self.gm, self.parameter)
