def add_likelihood(self, likelihood):
"""
Add likelihood distribution corresponding to all features in the
dataset.
"""
for n, f in enumerate(self.features):
fname = "%s:%s" % (self.name, xml.valid_id(f))
attribs = {
"id": "featureLikelihood:%s" % fname,
"spec": "TreeLikelihood",
"useAmbiguities": "true"
}
if self.pruned:
distribution = xml.distribution(likelihood, attrib=attribs)
# Create pruned tree
tree_id = "Tree.t:prunedBeastlingTree.%s" % fname
tree = xml.tree(distribution,
id=tree_id,
spec="beast.evolution.tree.PrunedTree",
quickshortcut="true",
attrib={'assert': "false"})
xml.tree(tree, idref="Tree.t:beastlingTree")
xml.alignment(tree, idref="pruned_data_%s" % fname)
# Create pruned branchrate
self.clock.add_pruned_branchrate_model(distribution, fname,
tree_id)
else:
attribs[
"branchRateModel"] = "@%s" % self.clock.branchrate_model_id
attribs["tree"] = "@Tree.t:beastlingTree"
distribution = xml.distribution(likelihood, attrib=attribs)
if f in self.reconstruct:
# Use a different likelihood spec (also depending on whether
# the whole tree is reconstructed, or only some nodes)
if self.treewide_reconstruction:
distribution.attrib[
"spec"] = "AncestralStateTreeLikelihood"
self.treedata.append(attribs["id"])
distribution.attrib["tag"] = f
else:
distribution.attrib[
"spec"] = "lucl.beast.statereconstruction.AncestralStatesLogger"
distribution.attrib["value"] = " ".join(
self.pattern_names(f))
for label in self.reconstruct_at:
langs = self.config.language_group(label)
self.beastxml.add_taxon_set(distribution, label, langs)
self.metadata.append(attribs["id"])
distribution.attrib["useAmbiguities"] = "false"
# Sitemodel
self.add_sitemodel(distribution, f, fname)
# Data
self.add_feature_data(distribution, n, f, fname)
def add_monophyly_constraints(self):
"""
Add monophyly constraints to prior distribution.
"""
if self.config.languages.monophyly:
xml.distribution(
self.prior,
id="constraints",
spec="beast.math.distributions.MultiMonophyleticConstraint",
tree="@{:}".format(self.config.treeprior.tree_id),
newick=self.config.languages.monophyly_newick,
)
def add_calibrations(self):
"""
Add timing calibrations to prior distribution.
"""
# This itertools.cchain is a bit ugly, I wonder if we can get away with sticking them all in one list...
for clade, cal in sorted(
itertools.chain(self.config.calibrations.items(),
self.config.tip_calibrations.items())):
# Don't add an MRCA cal for point calibrations, those only exist to
# cause the initial tip height to be set
if cal.dist == "point":
continue
# BEAST's logcombiner chokes on spaces...
clade = clade.replace(" ", "_")
# Create MRCAPrior node
attribs = {}
attribs["id"] = clade + "MRCA"
attribs["monophyletic"] = "true"
attribs["spec"] = "beast.math.distributions.MRCAPrior"
attribs["tree"] = "@{:}".format(self.config.treeprior.tree_id)
if cal.originate:
attribs["useOriginate"] = "true"
elif len(
cal.langs
) == 1: # If there's only 1 lang and it's not an originate cal, it must be a tip cal
attribs["tipsonly"] = "true"
cal_prior = xml.distribution(self.prior, attrib=attribs)
# Create "taxonset" param for MRCAPrior
self.add_taxon_set(
cal_prior, clade[:-len("_originate")]
if clade.endswith("_originate") else clade, cal.langs)
cal.generate_xml_element(cal_prior)
def add_likelihood(self):
"""
Add all likelihood distribution elements.
"""
self.likelihood = xml.distribution(self.posterior,
id="likelihood",
spec="util.CompoundDistribution")
for model in self.config.all_models:
model.add_likelihood(self.likelihood)
def add_distributions(self):
"""
Add all probability distributions under the <run> element.
"""
self.posterior = xml.distribution(self.run,
id="posterior",
spec="util.CompoundDistribution")
self.add_prior()
self.add_likelihood()
def add_prior(self, beastxml):
"""Add a (calibrated) birth-death tree prior."""
# Tree prior
attribs = {}
attribs["id"] = "BirthDeathModel.t:beastlingTree"
attribs["tree"] = "@Tree.t:beastlingTree"
attribs[
"spec"] = "beast.evolution.speciation.BirthDeathGernhard08Model"
attribs["birthDiffRate"] = "@birthRate.t:beastlingTree"
attribs["relativeDeathRate"] = "@deathRate.t:beastlingTree"
attribs["sampleProbability"] = "@sampling.t:beastlingTree"
attribs[
"type"] = "unscaled" #TODO: Someone dropped the "restricted" type here, which does not exist.
xml.distribution(beastxml.prior, attrib=attribs)
# Birth rate prior
attribs = {}
attribs["id"] = "BirthRatePrior.t:beastlingTree"
attribs["name"] = "distribution"
attribs["x"] = "@birthRate.t:beastlingTree"
sub_prior = xml.prior(beastxml.prior, attrib=attribs)
xml.Uniform(sub_prior, id="Uniform.0", name="distr", upper="Infinity")
# Relative death rate prior
attribs = {}
attribs["id"] = "relativeDeathRatePrior.t:beastlingTree"
attribs["name"] = "distribution"
attribs["x"] = "@deathRate.t:beastlingTree"
sub_prior = xml.prior(beastxml.prior, attrib=attribs)
xml.Uniform(sub_prior, id="Uniform.1", name="distr", upper="Infinity")
# Sample probability prior
attribs = {}
attribs["id"] = "samplingPrior.t:beastlingTree"
attribs["name"] = "distribution"
attribs["x"] = "@sampling.t:beastlingTree"
sub_prior = xml.prior(beastxml.prior, attrib=attribs)
xml.Uniform(sub_prior,
id="Uniform.3",
name="distr",
lower="0",
upper="1")
def add_prior(self, beastxml):
"""
Add Yule birth-process tree prior.
"""
# Tree prior
## Decide whether to use the standard Yule or the fancy calibrated one
if len(beastxml.config.calibrations) == 1:
yule = "calibrated"
elif len(beastxml.config.calibrations) == 2:
# Two calibrations can be handled by the calibrated Yule if they
# are nested
langs1, langs2 = [
c.langs for c in beastxml.config.calibrations.values()
]
if len(set(langs1) & set(langs2)) in (len(langs1), len(langs2)):
yule = "calibrated"
else:
yule = "standard"
else:
yule = "standard"
attribs = {}
attribs["id"] = "YuleModel.t:beastlingTree"
attribs["tree"] = "@Tree.t:beastlingTree"
if yule == "standard":
attribs["spec"] = "beast.evolution.speciation.YuleModel"
attribs["birthDiffRate"] = "@birthRate.t:beastlingTree"
if "root" in beastxml.config.calibrations:
attribs["conditionalOnRoot"] = "true"
elif yule == "calibrated":
attribs["spec"] = "beast.evolution.speciation.CalibratedYuleModel"
attribs["birthRate"] = "@birthRate.t:beastlingTree"
xml.distribution(beastxml.prior, attrib=attribs)
# Birth rate prior
sub_prior = xml.prior(beastxml.prior,
id="YuleBirthRatePrior.t:beastlingTree",
name="distribution",
x="@birthRate.t:beastlingTree")
xml.Uniform(sub_prior, id="Uniform.0", name="distr", upper="Infinity")
def add_prior(self, beastxml):
"""Add a Yule tree prior."""
coalescent = xml.distribution(beastxml.prior,
id="Coalescent.t:beastlingTree",
spec="Coalescent")
popmod = xml.populationModel(coalescent,
id="ConstantPopulation:beastlingTree",
spec="ConstantPopulation")
xml.parameter(popmod, idref="popSize.t:beastlingTree", name="popSize")
xml.treeIntervals(coalescent,
id="TreeIntervals",
spec="TreeIntervals",
tree="@Tree.t:beastlingTree")
def add_prior(self):
"""
Add all prior distribution elements.
"""
self.prior = xml.distribution(self.posterior,
id="prior",
spec="util.CompoundDistribution")
self.add_monophyly_constraints()
self.add_calibrations()
self.config.treeprior.add_prior(self)
for clock in self.config.clocks:
clock.add_prior(self.prior)
for model in self.config.all_models:
model.add_prior(self.prior)
def add_single_sitemodel_likelihood(self, likelihood):
attribs = {
"id": "DataLikelihood:%s" % self.name,
"spec": "TreeLikelihood",
"useAmbiguities": "true",
"branchRateModel": "@%s" % self.clock.branchrate_model_id,
"tree": "@Tree.t:beastlingTree",
}
distribution = xml.distribution(likelihood, attrib=attribs)
if not self.reconstruct:
pass
elif set(self.reconstruct) >= set(self.features):
# Use a different likelihood spec (also depending on whether
# the whole tree is reconstructed, or only some nodes)
if self.treewide_reconstruction:
distribution.attrib["spec"] = "ancestralstatetreelikelihood"
self.treedata.append(attribs["id"])
distribution.attrib["tag"] = f
else:
distribution.attrib[
"spec"] = "lucl.beast.statereconstruction.ancestralstateslogger"
distribution.attrib["value"] = " ".join(self.pattern_names(f))
for label in self.reconstruct_at:
langs = self.config.language_group(label)
self.beastxml.add_taxon_set(distribution, label, langs)
self.metadata.append(attribs["id"])
distribution.attrib["useAmbiguities"] = "false"
else:
raise NotImplementedError(
"The model {:} is a binarised model with a single site "
"model, so it uses a global likelihood. Reconstructing "
"only a subset of features is not supported.".format(
self.name))
self.add_sitemodel(distribution, None, None)
data = xml.data(distribution,
id="filtered_data_%s" % self.name,
spec="FilteredAlignment",
data="@data_%s" % self.name,
filter="-")
if self.recoded:
data.set("ascertained", "true")
data.set("excludefrom", "0")
if self.ascertained:
data.set("excludeto", "2")
else:
data.set("excludeto", "1")
data.append(self.get_userdatatype(None, None))
def add_single_sitemodel_likelihood(self, likelihood):
attribs = {
"id": "DataLikelihood:%s" % self.name,
"spec": "TreeLikelihood",
"useAmbiguities": "true",
"branchRateModel": "@%s" % self.clock.branchrate_model_id,
"tree": "@Tree.t:beastlingTree",
}
distribution = xml.distribution(likelihood, attrib=attribs)
self.add_sitemodel(distribution, None, None)
data = xml.data(distribution,
id="filtered_data_%s" % self.name,
spec="FilteredAlignment",
data="@data_%s" % self.name,
filter="-")
if self.recoded:
data.set("ascertained", "true")
data.set("excludefrom", "0")
if self.ascertained:
data.set("excludeto", "2")
else:
data.set("excludeto", "1")
data.append(self.get_userdatatype(None, None))
def add_likelihood(self, likelihood):
"""
Add likelihood distribution corresponding to all features in the
dataset.
"""
attribs = {
"id": "sphericalGeographyLikelihood",
"tree": "@Tree.t:beastlingTree",
"logAverage": "true",
"scale": "false",
"location": "@location.geo"
}
# Use appropriate Likelihood implementation depending
# upon presence/absence of sampled locations
if self.sampling_points:
attribs[
"spec"] = "sphericalGeo.ApproxMultivariateTraitLikelihoodF2"
else:
attribs["spec"] = "sphericalGeo.ApproxMultivariateTraitLikelihood"
distribution = xml.distribution(likelihood, attrib=attribs)
# Add data first, as this may trigger the creation of additional
# sampling points (for langs with missing locations data)
self.add_data(distribution)
# Now add geopriors
if self.sampling_points:
multi = xml.multiGeoprior(distribution,
id="multiGeoPrior",
spec="sphericalGeo.MultiGeoPrior",
tree="@Tree.t:beastlingTree",
newick="")
for clade in self.sampling_points:
# Get languages in clade
if clade == "root":
langs = self.config.languages.languages
else:
langs = self.config.language_group(clade)
if not langs:
continue
# Add the geo prior, which will trigger sampling
geoprior = xml.geoprior(multi,
id="%s.geoPrior" % clade,
spec="sphericalGeo.GeoPrior",
location="@location.geo",
tree="@Tree.t:beastlingTree")
if len(langs) > 1:
self.beastxml.add_taxon_set(geoprior, "%s.geo" % clade,
langs)
else:
xml.taxon(geoprior, idref=list(langs)[0])
# Drop back to F, not F2, so singletons can be sampled
distribution.set(
"spec",
"sphericalGeo.ApproxMultivariateTraitLikelihoodF")
# Also add the KML file if we have an actual constraint
if clade in self.geo_priors:
xml.region(geoprior,
spec="sphericalGeo.region.KMLRegion",
kml=self.geo_priors[clade])
self.add_sitemodel(distribution)
xml.branchRateModel(distribution, idref=self.clock.branchrate_model_id)