def sampling_dishes(self, iteration):
""" sample a topic of each table
"""
# need to check this function
for doc in self.doc_list:
tables = doc.get_tables()
for table in tables:
tableWords = doc.tableWords[table]
old_topic = doc.tableTopic[table]
# remove current topic of table
self.table_assigned_topics[old_topic] -= 1
if self.table_assigned_topics[old_topic] == 0:
del self.table_assigned_topics[old_topic]
for wordNo, counts in tableWords.iteritems():
self.word_topic.decrease(wordNo, old_topic, counts)
topic_prob = dict()
for topicNo in self.table_assigned_topics.keys():
topic_prob[topicNo] = np.log(
self.table_assigned_topics[topicNo]
) + self.word_topic.get_multiword_log_conditional(
tableWords, topicNo, self.eta)
new_topic_no = self.get_new_topic()
topic_prob[new_topic_no] = np.log(
self.alpha
) + self.word_topic.get_multiword_log_conditional(
tableWords, new_topic_no, self.eta)
topic_prob = log_normalize(topic_prob)
new_topic = sampling_from_dict(topic_prob)
doc.tableTopic[table] = new_topic
# if a new topic is chosen
if new_topic == new_topic_no:
self.table_assigned_topics[new_topic] = 0
self.table_assigned_topics[new_topic] += 1
for wordNo, counts in tableWords.iteritems():
self.word_topic.increase(wordNo, new_topic, counts)
def sampling_dishes(self, iteration):
""" sample a topic of each table
"""
# need to check this function
for doc in self.doc_list:
tables = doc.get_tables()
for table in tables:
tableWords = doc.tableWords[table]
old_topic = doc.tableTopic[table]
# remove current topic of table
self.table_assigned_topics[old_topic] -= 1
if self.table_assigned_topics[old_topic] == 0:
del self.table_assigned_topics[old_topic]
for wordNo, counts in tableWords.iteritems():
self.word_topic.decrease(wordNo, old_topic, counts)
topic_prob = dict()
for topicNo in self.table_assigned_topics.keys():
topic_prob[topicNo] = np.log(self.table_assigned_topics[topicNo]) + self.word_topic.get_multiword_log_conditional(tableWords, topicNo, self.eta)
new_topic_no = self.get_new_topic()
topic_prob[new_topic_no] = np.log(self.alpha) + self.word_topic.get_multiword_log_conditional(tableWords, new_topic_no, self.eta)
topic_prob = log_normalize(topic_prob)
new_topic = sampling_from_dict(topic_prob)
doc.tableTopic[table] = new_topic
# if a new topic is chosen
if new_topic == new_topic_no:
self.table_assigned_topics[new_topic] = 0
self.table_assigned_topics[new_topic] += 1
for wordNo, counts in tableWords.iteritems():
self.word_topic.increase(wordNo, new_topic, counts)
def sampling_tables(self, iteration):
""" iterate a corpus and sample a table of each word token
Keyword arguments:
iteration -- current iteration count
"""
for doc in self.doc_list:
doc_length = doc.get_doc_length() - 1
for word in doc.word_list:
wordNo = word.wordNo
# remove current word from assigned table
if iteration != 0:
old_table = word.tableNo
old_topic = doc.tableTopic[old_table]
doc.remove_word_from_table(wordNo, old_table)
if doc.get_table_sum(old_table) == 0:
self.table_assigned_topics[old_topic] -= 1
self.total_table -= 1
if self.table_assigned_topics[old_topic] == 0:
del self.table_assigned_topics[old_topic]
self.word_topic.decrease(wordNo, old_topic)
# compute conditional for each table, topic
tables = doc.get_tables()
topic_prob = dict()
for topicNo in self.word_topic.get_topics():
topic_prob[topicNo] = self.word_topic.get_conditional(wordNo,topicNo,self.eta)
new_topic_no = self.word_topic.get_new_topic()
topic_prob[new_topic_no] = self.word_topic.get_conditional(wordNo,new_topic_no, self.eta)
table_prob = dict()
for tableNo in tables:
table_prob[tableNo] = topic_prob[doc.tableTopic[tableNo]] * (doc.tableSum[tableNo])/(doc_length + self.beta)
new_table_no = doc.get_new_table()
new_table_prob = 0
new_table_dict = dict()
for topicNo in topic_prob.keys():
if self.table_assigned_topics.has_key(topicNo):
prob = (self.table_assigned_topics[topicNo])/(self.total_table + self.alpha) * topic_prob[topicNo]
else:
prob = self.alpha/(self.total_table + self.alpha) * topic_prob[topicNo]
new_table_prob += prob
new_table_dict[topicNo] = prob
table_prob[new_table_no] = new_table_prob * self.beta / (doc_length + self.beta)
new_table = sampling_from_dict(table_prob)
# if a new table is chosen
if new_table == new_table_no:
new_topic_of_new_table = sampling_from_dict(new_table_dict)
self.total_table += 1
# if a new topic is chosen for the new table
if new_topic_of_new_table == new_topic_no:
self.table_assigned_topics[new_topic_of_new_table] = 0
self.table_assigned_topics[new_topic_of_new_table] += 1
doc.tableTopic[new_table] = new_topic_of_new_table
word.tableNo = new_table
doc.add_word_to_table(wordNo, new_table)
self.word_topic.increase(wordNo, doc.tableTopic[new_table])
def sampling_tables(self, iteration):
""" iterate a corpus and sample a table of each word token
Keyword arguments:
iteration -- current iteration count
"""
for doc in self.doc_list:
doc_length = doc.get_doc_length() - 1
for word in doc.word_list:
wordNo = word.wordNo
# remove current word from assigned table
if iteration != 0:
old_table = word.tableNo
old_topic = doc.tableTopic[old_table]
doc.remove_word_from_table(wordNo, old_table)
if doc.get_table_sum(old_table) == 0:
self.table_assigned_topics[old_topic] -= 1
self.total_table -= 1
if self.table_assigned_topics[old_topic] == 0:
del self.table_assigned_topics[old_topic]
self.word_topic.decrease(wordNo, old_topic)
# compute conditional for each table, topic
tables = doc.get_tables()
topic_prob = dict()
for topicNo in self.word_topic.get_topics():
topic_prob[topicNo] = self.word_topic.get_conditional(
wordNo, topicNo, self.eta)
new_topic_no = self.word_topic.get_new_topic()
topic_prob[new_topic_no] = self.word_topic.get_conditional(
wordNo, new_topic_no, self.eta)
table_prob = dict()
for tableNo in tables:
table_prob[tableNo] = topic_prob[
doc.tableTopic[tableNo]] * (doc.tableSum[tableNo]) / (
doc_length + self.beta)
new_table_no = doc.get_new_table()
new_table_prob = 0
new_table_dict = dict()
for topicNo in topic_prob.keys():
if self.table_assigned_topics.has_key(topicNo):
prob = (self.table_assigned_topics[topicNo]) / (
self.total_table +
self.alpha) * topic_prob[topicNo]
else:
prob = self.alpha / (self.total_table +
self.alpha) * topic_prob[topicNo]
new_table_prob += prob
new_table_dict[topicNo] = prob
table_prob[new_table_no] = new_table_prob * self.beta / (
doc_length + self.beta)
new_table = sampling_from_dict(table_prob)
# if a new table is chosen
if new_table == new_table_no:
new_topic_of_new_table = sampling_from_dict(new_table_dict)
self.total_table += 1
# if a new topic is chosen for the new table
if new_topic_of_new_table == new_topic_no:
self.table_assigned_topics[new_topic_of_new_table] = 0
self.table_assigned_topics[new_topic_of_new_table] += 1
doc.tableTopic[new_table] = new_topic_of_new_table
word.tableNo = new_table
doc.add_word_to_table(wordNo, new_table)
self.word_topic.increase(wordNo, doc.tableTopic[new_table])
