class MorphoLDA(TopicModel):
def __init__(self, n_topics, n_docs, pattern_base, stem_base, analyses):
super(MorphoLDA, self).__init__(n_topics)
self.alpha = GammaPrior(1.0, 1.0, 1.0)
self.document_topic = [DirichletMultinomial(n_topics, self.alpha) for _ in xrange(n_docs)]
self.stem_base = stem_base
self.pattern_model = PYP(pattern_base, PYPPrior(1.0, 1.0, 1.0, 1.0, 0.1, 1.0)) # G_p
def make_topic_word():
stem_model = PYP(stem_base, PYPPrior(1.0, 1.0, 1.0, 1.0, 0.1, 1.0)) # G_s
mp = MorphoProcess(stem_model, self.pattern_model, analyses)
return PYP(mp, PYPPrior(1.0, 1.0, 1.0, 1.0, 0.1, 1.0)) # G_w
self.topic_word = [make_topic_word() for _ in xrange(n_topics)]
def log_likelihood(self):
return (sum(d.log_likelihood() for d in self.document_topic)
+ self.alpha.log_likelihood()
+ sum(topic.log_likelihood() # G_w
+ topic.prior.log_likelihood() # d_w, T_w
+ topic.base.stem_model.log_likelihood() # G_s
+ topic.base.stem_model.prior.log_likelihood() # d_s, T_s
for topic in self.topic_word)
+ self.pattern_model.log_likelihood(full=True) # G_p + ...
+ self.stem_base.log_likelihood(full=True)) # G_s^0 + ...
def resample_hyperparemeters(self, n_iter):
ar = stuple((0, 0))
logging.info('Resampling doc-topic hyperparameters')
ar += self.alpha.resample(n_iter)
logging.info('Resampling stem base / pattern model hyperparameters')
ar += self.pattern_model.resample_hyperparemeters(n_iter) # G_p
ar += self.pattern_model.base.resample_hyperparemeters(n_iter) # G_p^0
ar += self.stem_base.resample_hyperparemeters(n_iter) # G_s^0
logging.info('Resampling all topic-word PYP hyperparameters')
for topic in self.topic_word:
ar += topic.resample_hyperparemeters(n_iter) # G_w
ar += topic.base.stem_model.resample_hyperparemeters(n_iter) # G_s
return ar
def __repr__(self):
return ('MorphoLDA(#topics={self.n_topics} '
'| alpha={self.alpha}, beta=PYP(base=MP(stem ~ PYP(base={self.stem_base}); '
'pattern ~ {self.pattern_model})))').format(self=self)
def __init__(self, n_topics, n_docs, pattern_base, stem_base, analyses):
super(MorphoLDA, self).__init__(n_topics)
self.alpha = GammaPrior(1.0, 1.0, 1.0)
self.document_topic = [DirichletMultinomial(n_topics, self.alpha) for _ in xrange(n_docs)]
self.stem_base = stem_base
self.pattern_model = PYP(pattern_base, PYPPrior(1.0, 1.0, 1.0, 1.0, 0.1, 1.0)) # G_p
def make_topic_word():
stem_model = PYP(stem_base, PYPPrior(1.0, 1.0, 1.0, 1.0, 0.1, 1.0)) # G_s
mp = MorphoProcess(stem_model, self.pattern_model, analyses)
return PYP(mp, PYPPrior(1.0, 1.0, 1.0, 1.0, 0.1, 1.0)) # G_w
self.topic_word = [make_topic_word() for _ in xrange(n_topics)]
def __getitem__(self, ctx):
""" create a new PYP if the context has not been seen """
if ctx not in self.models:
base = (self.backoff if self.order == 1 else BackoffBase(self.backoff, ctx[1:]))
return PYP(base, self.prior)
return self.models[ctx]
