def estimate_DiscreteMixture(histo, *args, **kargs):
""" Estimate a finite mixture of discrete distributions
:Parameters:
* histo (histogram, mixture_data, convolution_data, compound_data),
* distributions (list) : a list of distribution object
or distribution label(string) : 'B', 'NB', 'U', 'P', ...
* unknown (string): type of unknown distribution: "Sum" or "Elementary".
:Keywords:
* MinInfBound (int): lower bound to the range of possible values (0 -default- or 1). \
This optional argument cannot be used in conjunction \
with the optional argument InitialDistribution.
* InfBoundStatus (string): shifting or not of the distribution: "Free" (default value) or "Fixed".
* DistInfBoundStatus (string): shifting or not of the subsequent components of \
the mixture: "Free" (default value) or "Fixed".
* NbComponent (string): estimation of the number of components of the mixture: \
"Fixed" (default value) or "Estimated". Le number of estimated \
components is comprised between\
1 and a maximum number which is given by the number of specified \
parametric distributions in the mandatory arguments \
(all of these distributions are assumed to be unknown).
* Penalty (string): type of Penalty function for model selection: \
"AIC" (Akaike Information Criterion), \
"AICc" (corrected Akaike Information Criterion) \
"BIC" (Bayesian Information Criterion - default value). \
"BICc" (corrected Bayesian Information Criterion). \
This optional argument can only be used if the optional argument
NbComponent is set at "Estimated".
:Examples:
.. doctest::
:options: +SKIP
>>> estimate_DiscreteMixture(histo, "MIXTURE", "B", dist,...,,
MinInfBound=1, InfBoundStatus="Fixed",
DistInfBoundStatus="Fixed")
>>> estimate_DiscreteMixture(histo, "MIXTURE", "B", "NB",...,,
MinInfBound=1, InfBoundStatus="Fixed",
DistInfBoundStatus="Fixed",
NbComponent="Estimated", Penalty="AIC")
>>> Estimate(histo, "MIXTURE", "B", dist, MinInfBound=1, InfBoundStatus="Fixed",
DistInfBoundStatus="Fixed")
>>> Estimate(histo, "MIXTURE", "B", "NB",
MinInfBound=1, InfBoundStatus="Fixed",
DistInfBoundStatus="Fixed",
NbComponent="Estimated", Penalty="AIC")
"""
#alias
#error.CheckArgumentsLength(args, 1, 1)
# get user arguments
# list of distributions can be either a list or several arguments
# e.g.: estimate_DiscreteMixture(["B","B"]) or estimate_DiscreteMixture("B", "B")
if len(args)==1 and type(args[0])==list:
distributions = args[0]
else:
distributions = list(args)
InfBoundStatus = kargs.get("InfBoundStatus","Free")
DistInfBoundStatus = kargs.get("DistInfBoundStatus", "Free")
NbComponent = kargs.get("NbComponent", "Fixed")
MinInfBound = kargs.get("MinInfBound", 0)
Penalty = error.ParseKargs(kargs, "Penalty", "AIC",
likelihood_penalty_type)
#should be before the conversion to booleans
error.CheckType([MinInfBound, InfBoundStatus, DistInfBoundStatus,
NbComponent, Penalty],
[int, str, str, str, LikelihoodPenaltyType])
# transform into boolean when needed
InfBoundStatus = bool(InfBoundStatus == "Free")
DistInfBoundStatus = bool(DistInfBoundStatus == "Free")
NbComponent = bool(NbComponent == "Estimated")
estimate = [] # list of bool
pcomponent = [] # list of distribution
ident = [] # list of distribution identifier
# Parse list of distribution that could be defined by a distribution,
# compound, mixture, convolution or simplya string such as "B",
# "Mixture", ...
for dist in distributions:
if isinstance(dist, str):
dist_authorised = ["BINOMIAL", "B", "POISSON",
"P", "NB", "NEGATIVE_BINOMIAL"]
if dist not in dist_authorised:
raise ValueError("""If distribution is a string, then it
must be in %s. You provided %s"""
% (dist_authorised, dist))
#todo: check that poisson is allowed
pcomponent.append(_DiscreteParametric(0, dist_type[dist]))
ident.append(dist_type[dist])
estimate.append(True)
elif isinstance(dist, _DiscreteParametricModel):
pcomponent.append(_DiscreteParametric(dist))
ident.append(None)
estimate.append(False)
elif type(dist) in [_DiscreteMixture, _Convolution, _Compound]:
pcomponent.append(_Distribution(dist))
ident.append(None)
estimate.append(False)
else:
raise ValueError("""In the case of a MIXTURE estimation,
argument related to distributions must be either string, or
Distribution, Mixture, Convolution, Compound. %s provided"""
% dist)
# check parameters
if not NbComponent and Penalty:
raise TypeError("""
Penalty can only be used with NbComponent set to 'Estimated'""")
if not NbComponent: # "FIXED"
imixt = _DiscreteMixture(pcomponent)
ret = histo.discrete_mixture_estimation1(imixt, estimate, MinInfBound,
InfBoundStatus, DistInfBoundStatus)
return ret
else: # "ESTIMATED"
ret = histo.discrete_mixture_estimation2(ident, MinInfBound, InfBoundStatus,
DistInfBoundStatus, Penalty)
return ret
def estimate_DiscreteMixture(histo, *args, **kargs):
""" Estimate a finite mixture of discrete distributions
:Parameters:
* histo (histogram, mixture_data, convolution_data, compound_data),
* distributions (list) : a list of distribution object
or distribution label(string) : 'B', 'NB', 'U', 'P', ...
* unknown (string): type of unknown distribution: "Sum" or "Elementary".
:Keywords:
* MinInfBound (int): lower bound to the range of possible values (0 -default- or 1). \
This optional argument cannot be used in conjunction \
with the optional argument InitialDistribution.
* InfBoundStatus (string): shifting or not of the distribution: "Free" (default value) or "Fixed".
* DistInfBoundStatus (string): shifting or not of the subsequent components of \
the mixture: "Free" (default value) or "Fixed".
* NbComponent (string): estimation of the number of components of the mixture: \
"Fixed" (default value) or "Estimated". Le number of estimated \
components is comprised between\
1 and a maximum number which is given by the number of specified \
parametric distributions in the mandatory arguments \
(all of these distributions are assumed to be unknown).
* Penalty (string): type of Penalty function for model selection: \
"AIC" (Akaike Information Criterion), \
"AICc" (corrected Akaike Information Criterion) \
"BIC" (Bayesian Information Criterion - default value). \
"BICc" (corrected Bayesian Information Criterion). \
This optional argument can only be used if the optional argument
NbComponent is set at "Estimated".
:Examples:
.. doctest::
:options: +SKIP
>>> estimate_DiscreteMixture(histo, "MIXTURE", "B", dist,...,,
MinInfBound=1, InfBoundStatus="Fixed",
DistInfBoundStatus="Fixed")
>>> estimate_DiscreteMixture(histo, "MIXTURE", "B", "NB",...,,
MinInfBound=1, InfBoundStatus="Fixed",
DistInfBoundStatus="Fixed",
NbComponent="Estimated", Penalty="AIC")
>>> Estimate(histo, "MIXTURE", "B", dist, MinInfBound=1, InfBoundStatus="Fixed",
DistInfBoundStatus="Fixed")
>>> Estimate(histo, "MIXTURE", "B", "NB",
MinInfBound=1, InfBoundStatus="Fixed",
DistInfBoundStatus="Fixed",
NbComponent="Estimated", Penalty="AIC")
"""
#alias
#error.CheckArgumentsLength(args, 1, 1)
# get user arguments
# list of distributions can be either a list or several arguments
# e.g.: estimate_DiscreteMixture(["B","B"]) or estimate_DiscreteMixture("B", "B")
if len(args) == 1 and type(args[0]) == list:
distributions = args[0]
else:
distributions = list(args)
InfBoundStatus = kargs.get("InfBoundStatus", "Free")
DistInfBoundStatus = kargs.get("DistInfBoundStatus", "Free")
NbComponent = kargs.get("NbComponent", "Fixed")
MinInfBound = kargs.get("MinInfBound", 0)
Penalty = error.ParseKargs(kargs, "Penalty", "AIC",
likelihood_penalty_type)
#should be before the conversion to booleans
error.CheckType([
MinInfBound, InfBoundStatus, DistInfBoundStatus, NbComponent,
Penalty
], [int, str, str, str, LikelihoodPenaltyType])
# transform into boolean when needed
InfBoundStatus = bool(InfBoundStatus == "Free")
DistInfBoundStatus = bool(DistInfBoundStatus == "Free")
NbComponent = bool(NbComponent == "Estimated")
estimate = [] # list of bool
pcomponent = [] # list of distribution
ident = [] # list of distribution identifier
# Parse list of distribution that could be defined by a distribution,
# compound, mixture, convolution or simplya string such as "B",
# "Mixture", ...
for dist in distributions:
if isinstance(dist, str):
dist_authorised = [
"BINOMIAL", "B", "POISSON", "P", "NB", "NEGATIVE_BINOMIAL"
]
if dist not in dist_authorised:
raise ValueError("""If distribution is a string, then it
must be in %s. You provided %s""" %
(dist_authorised, dist))
#todo: check that poisson is allowed
pcomponent.append(_DiscreteParametric(0, dist_type[dist]))
ident.append(dist_type[dist])
estimate.append(True)
elif isinstance(dist, _DiscreteParametricModel):
pcomponent.append(_DiscreteParametric(dist))
ident.append(None)
estimate.append(False)
elif type(dist) in [_DiscreteMixture, _Convolution, _Compound]:
pcomponent.append(_Distribution(dist))
ident.append(None)
estimate.append(False)
else:
raise ValueError("""In the case of a MIXTURE estimation,
argument related to distributions must be either string, or
Distribution, Mixture, Convolution, Compound. %s provided""" %
dist)
# check parameters
if not NbComponent and Penalty:
raise TypeError("""
Penalty can only be used with NbComponent set to 'Estimated'""")
if not NbComponent: # "FIXED"
imixt = _DiscreteMixture(pcomponent)
ret = histo.discrete_mixture_estimation1(imixt, estimate,
MinInfBound,
InfBoundStatus,
DistInfBoundStatus)
return ret
else: # "ESTIMATED"
ret = histo.discrete_mixture_estimation2(ident, MinInfBound,
InfBoundStatus,
DistInfBoundStatus,
Penalty)
return ret
def Mixture(*args):
"""Construction of a mixture of distributions from elementary distributions
and associated weights or from an ASCII file.
A mixture is a parametric model of classification where each elementary
distribution or component represents a class with its associated weight.
:Parameters:
* `weight1`, `weight2`, ... (float) - weights of each component.
These weights should sum to one (they constitute a discrete
distribution).
* `dist1`, `dist2`, ... (`_DiscreteParametricModel`, `_DiscreteMixture`, `_Convolution`,
`_Compound`) elementary distributions (or components).
* `filename` (string) -
:Returns:
If the construction succeeds, an object of type mixture is returned,
otherwise no object is returned.
:Examples:
.. doctest::
:options: +SKIP
>>> Mixture(weight1, dist1, weight2, dist2,...)
>>> Mixture(filename)
.. seealso::
:func:`~openalea.stat_tool.output.Save`,
:func:`~openalea.stat_tool.estimate.Estimate`,
:func:`~openalea.stat_tool.simulate.Simulate`.
"""
error.CheckArgumentsLength(args, 1)
types = [
_DiscreteParametricModel, _DiscreteMixture, _Compound, _Convolution
]
# filename
if (len(args) == 1):
error.CheckType([args[0]], [str], arg_id=[1])
result = _DiscreteMixture(args[0])
# build list of weights and distributions
else:
nb_param = len(args)
if ((nb_param % 2) != 0):
raise TypeError("Number of parameters must be pair")
# Extract weights ands distributions
weights = []
dists = []
for i in xrange(nb_param / 2):
weights.append(args[i * 2])
error.CheckType([args[i * 2 + 1]], [types], arg_id=[i * 2 + 1])
error.CheckType([args[i * 2]], [float], arg_id=[i * 2])
#dists.append(_Distribution(args[i * 2 + 1]))
dists.append((args[i * 2 + 1]))
result = _DiscreteMixture(weights, dists)
return result
def Mixture(*args):
"""Construction of a mixture of distributions from elementary distributions
and associated weights or from an ASCII file.
A mixture is a parametric model of classification where each elementary
distribution or component represents a class with its associated weight.
:Parameters:
* `weight1`, `weight2`, ... (float) - weights of each component.
These weights should sum to one (they constitute a discrete
distribution).
* `dist1`, `dist2`, ... (`_DiscreteParametricModel`, `_DiscreteMixture`, `_Convolution`,
`_Compound`) elementary distributions (or components).
* `filename` (string) -
:Returns:
If the construction succeeds, an object of type mixture is returned,
otherwise no object is returned.
:Examples:
.. doctest::
:options: +SKIP
>>> Mixture(weight1, dist1, weight2, dist2,...)
>>> Mixture(filename)
.. seealso::
:func:`~openalea.stat_tool.output.Save`,
:func:`~openalea.stat_tool.estimate.Estimate`,
:func:`~openalea.stat_tool.simulate.Simulate`.
"""
error.CheckArgumentsLength(args, 1)
types = [_DiscreteParametricModel, _DiscreteMixture, _Compound, _Convolution]
# filename
if (len(args) == 1):
error.CheckType([args[0]], [str], arg_id=[1])
result = _DiscreteMixture(args[0])
# build list of weights and distributions
else:
nb_param = len(args)
if ((nb_param % 2) != 0):
raise TypeError("Number of parameters must be pair")
# Extract weights ands distributions
weights = []
dists = []
for i in xrange(nb_param / 2):
weights.append(args[i * 2])
error.CheckType([args[i * 2 + 1]], [types], arg_id=[i * 2 + 1])
error.CheckType([args[i * 2]], [float], arg_id=[i * 2])
#dists.append(_Distribution(args[i * 2 + 1]))
dists.append((args[i * 2 + 1]))
result = _DiscreteMixture(weights, dists)
return result