def add_substmodel(self, sitemodel, feature, fname):
key = self._get_substmodel_key(feature)
if key and key in self.subst_models:
substmodel = xml.substModel(sitemodel,
idref=self.subst_models[key])
else:
sm_id = "mk.s:%s" % fname
substmodel = xml.substModel(sitemodel,
id=sm_id,
spec="LewisMK",
datatype="@featureDataType.%s" % fname)
# Do empirical frequencies
# We don't need to do anything for uniform freqs
# as the implementation of LewisMK handles it
if self.frequencies == "empirical":
xml.frequencies(substmodel,
id="feature_freqs.s:%s" % fname,
spec="Frequencies",
data="@feature_data_%s" % fname)
elif self.frequencies == "approx":
xml.frequencies(
substmodel,
id="feature_freqs.s:%s" % fname,
spec="Frequencies",
frequencies=self._get_approx_freq_string(feature))
elif self.frequencies == "estimate":
xml.frequencies(substmodel,
id="feature_freqs.s:%s" % fname,
spec="Frequencies",
frequencies="@feature_freqs_param.s:%s" %
fname)
self.subst_models[key] = sm_id
return substmodel
def add_sitemodel(self, distribution):
site = xml.siteModel(distribution,
id="sphericalGeoSiteModel",
spec="SiteModel")
xml.substModel(site,
id="sphericalDiffusionSubstModel",
spec="sphericalGeo.SphericalDiffusionModel",
precision="@sphericalPrecision",
fast="true",
threshold="1")
def _add_substmodel(self, sitemodel, feature, name):
if self.share_params:
name = self.name
self.subst_model_id = "%s:pdcovarion.s" % name
subst_model_id = "%s:pdcovarion.s" % name
substmodel = xml.substModel(
sitemodel,
id=subst_model_id,
spec="BirthDeathCovarion2",
deathprob="@{:}:pdcovarion_death.s".format(name),
originLength="@{:s}:pdcovarion_origin.s".format(name),
switchRate="@{:}:pdcovarion_s.s".format(name))
# Numerical instability is an issue with this model, so we give the
# option of using a more robust method of computing eigenvectors.
if self.use_robust_eigensystem: # pragma: no cover
raise ValueError(
"Currently, Beast's pseudo-Dollo covarion model does not "
"support robust eigensystems.")
substmodel.set("eigenSystem", "beast.evolution.substitutionmodel.RobustEigenSystem")
# The "vfrequencies" parameter here is the frequencies
# of the *visible* states (present/absent) and should
# be based on the data (if we are doing an empirical
# analysis)
if self.frequencies == "estimate":
substmodel.set("vfrequencies","@%s:visiblefrequencies.s" % name)
else:
vfreq = xml.vfrequencies(
substmodel,
id="%s:visiblefrequencies.s" % name,
dimension="3",
spec="parameter.RealParameter")
if self.frequencies == "empirical": # pragma: no cover
raise ValueError("Dollo model {:} cannot derive empirical "
"frequencies from data".format(self.name))
else:
vfreq.text="0.94 0.05 0.01"
# These are the frequencies of the *hidden* states
# (fast / slow), and are just set to 50: 50. They could be estimated,
# in principle, but this seems to lead to serious instability problems
# so we don't expose that possibility to the user.
xml.parameter(
substmodel,
text="0.5 0.5",
id="%s: hiddenfrequencies.s" % name,
dimension="2",
name="hfrequencies",
lower="0.0",
upper="1.0")
def add_substmodel(self, sitemodel, feature, fname):
# If we're sharing one substmodel across all features and have already
# created it, just reference it and that's it
if self.subst_model_id:
sitemodel.set("substModel", "@%s" % self.subst_model_id)
return
# Otherwise, create a substmodel
name = self.name if self.share_params else fname
subst_model_id = "binaryCTMC.s:%s" % name
if self.share_params:
self.subst_model_id = subst_model_id
substmodel = xml.substModel(sitemodel,
id=subst_model_id,
spec="GeneralSubstitutionModel")
xml.parameter(substmodel,
text="1.0 1.0",
id="rates.s:%s" % name,
dimension=2,
estimate="false",
name="rates")
if self.frequencies == "estimate":
xml.frequencies(substmodel,
id="estimatedFrequencies.s:%s" % name,
spec="Frequencies",
frequencies="@freqs_param.s:%s" % name)
elif self.frequencies == "empirical":
attribs = {
"id": "empiricalFrequencies.s:%s" % name,
"spec": "Frequencies"
}
if self.share_params:
if self.single_sitemodel:
attribs["data"] = "@filtered_data_%s" % name
else:
attribs["frequencies"] = self.build_freq_str()
else:
attribs["data"] = "@feature_data_%s" % name
xml.frequencies(substmodel, attrib=attribs)
elif self.frequencies == "uniform":
xml.frequencies(substmodel,
text="0.5 0.5",
id="frequencies.s:%s" % name,
dimension="2",
spec="parameter.RealParameter")
def add_substmodel(self, sitemodel, feature, fname):
attribs = {
"id": "svs.s:%s"%fname,
"rateIndicator": "@rateIndicator.s:%s" % fname,
"rates": "@relativeGeoRates.s:%s" % fname,
"spec": "SVSGeneralSubstitutionModel"}
if not self.symmetric:
attribs['symmetric'] = 'false'
if self.use_robust_eigensystem:
attribs["eigenSystem"] = "beast.evolution.substitutionmodel.RobustEigenSystem"
substmodel = xml.substModel(sitemodel, **attribs)
attribs = {
"id": "feature_freqs.s:%s"%fname,
"spec": "Frequencies",
}
freq_string=None
if self.frequencies == "estimate":
attribs["frequencies"] = "@feature_freqs_param.s:%s"%fname
elif self.frequencies == "uniform":
freq_string = str(1.0/self.valuecounts[feature])
elif self.frequencies == "empirical":
#TODO: Do this in the BEAStly way
freqs = [
self.counts[feature].get(
self.unique_values[feature][v], 0)
for v in range(self.valuecounts[feature])]
norm = float(sum(freqs))
freqs = [f/norm for f in freqs]
# Sometimes, due to WALS oddities, there's a zero frequency, and that makes BEAST sad. So do some smoothing in these cases:
if 0 in freqs:
freqs = [0.1/self.valuecounts[feature] + 0.9*f for f in freqs]
norm = float(sum(freqs))
freq_string = " ".join([str(c/norm) for c in freqs])
else:
raise ValueError(
"Model BSVS does not recognize frequencies %r, "
"should be 'uniform' or 'empirical'." % self.frequencies)
freq = xml.frequencies(substmodel, **attribs)
if self.frequencies != "estimate":
xml.parameter(
freq,
text=freq_string,
dimension=self.valuecounts[feature],
id="feature_frequencies.s:%s" % fname,
name="frequencies")
def add_substmodel(self, sitemodel, feature, fname):
# If we're sharing one substmodel across all features and have already
# created it, just reference it and that's it
if self.share_params and self.subst_model_id:
sitemodel.set("substModel", "@%s" % self.subst_model_id)
return
# Otherwise, create a substmodel
name = self.name if self.share_params else fname
subst_model_id = "covarion.s:%s" % name
if self.share_params:
self.subst_model_id = subst_model_id
substmodel = xml.substModel(sitemodel,
id=subst_model_id,
spec="BinaryCovarion",
alpha="@covarion_alpha.s:%s" % name,
switchRate="@covarion_s.s:%s" % name)
# Numerical instability is an issue with this model, so we give the
# option of using a more robust method of computing eigenvectors.
if self.use_robust_eigensystem:
substmodel.set(
"eigenSystem",
"beast.evolution.substitutionmodel.RobustEigenSystem")
# The "vfrequencies" parameter here is the frequencies
# of the *visible* states (present/absent) and should
# be based on the data (if we are doing an empirical
# analysis)
if self.frequencies == "estimate":
substmodel.set("vfrequencies", "@freqs_param.s:%s" % name)
else:
vfreq = xml.vfrequencies(substmodel,
id="%s:visiblefrequencies.s" % name,
dimension="2",
spec="parameter.RealParameter")
if self.frequencies == "empirical":
if self.share_params:
vfreq.text = self.build_freq_str()
else:
vfreq.text = self.build_freq_str(feature)
else:
vfreq.text = "0.5 0.5"
# These are the frequencies of the *hidden* states
# (fast / slow), and are just set to 50:50. They could be estimated,
# in principle, but this seems to lead to serious instability problems
# so we don't expose that possibility to the user.
xml.parameter(substmodel,
text="0.5 0.5",
id="%s:hiddenfrequencies.s" % name,
dimension="2",
lower="0.0",
name="hfrequencies",
upper="1.0")
# Dummy frequencies - these do nothing and are required
# to stop the BinaryCovarion model complaining that the
# "frequencies" input is not specified, which is
# inherited behaviour from GeneralSubstitutionModel
# which probably should have been overridden...
xml.frequencies(substmodel,
id="%s:dummyfrequences.s" % name,
spec="Frequencies",
frequencies="0.5 0.5")