def generate_unique_trees(grammar, start='START', N=1000):
"""
Yield a bunch of unique trees, produced from the grammar
"""
for _ in break_ctrlc(range(N)):
t = grammar.generate(start)
yield t
class MyHypothesis(GaussianLikelihood, LOTHypothesis):
def __init__(self, **kwargs):
LOTHypothesis.__init__(self, grammar=grammar, **kwargs)
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Main
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
if __name__ == "__main__":
from LOTlib3 import break_ctrlc
from LOTlib3.Miscellaneous import qq
from LOTlib3.TopN import TopN
from LOTlib3.Samplers.MetropolisHastings import MetropolisHastingsSampler
h0 = MyHypothesis()
data = make_data()
top = TopN(N=10)
thin = 100
for i, h in enumerate(break_ctrlc(MetropolisHastingsSampler(h0, data))):
top << h
if i % thin == 0:
print("#", i, h.posterior_score, h.prior, h.likelihood, qq(h))
for h in top:
print(h.posterior_score, h.prior, h.likelihood, qq(h))
class PriorDirichletDistribution(DirichletDistribution):
""" propose from the prior (useful for debugging)
"""
def propose(self):
if len(self.value) == 1:
return PriorDirichletDistribution(
value=self.value,
alpha=self.value), 0.0 # handle singleton rules
ret = PriorDirichletDistribution(value=numpy.random.dirichlet(
self.alpha),
alpha=self.alpha)
fb = dirichlet.logpdf(ret.value, self.alpha) - dirichlet.logpdf(
self.value, self.alpha)
return ret, fb
class BetaDistribution(DirichletDistribution):
def __init__(self, a=1, b=1, **kwargs):
DirichletDistribution.__init__(self, alpha=[a, b], **kwargs)
if __name__ == "__main__":
from LOTlib3.Inference.Samplers.MetropolisHastings import MHSampler
h0 = BetaDistribution(1, 2)
for h in break_ctrlc(MHSampler(h0, [])):
print(h)