def test_topics(self):
naive = {"Health": 0, "Sports": 0}
for pi in self.postinteractions_set:
for (topic, weight) in self._interactions_weights(pi):
naive[topic] += weight
topics = {topic: utils.atan_norm(score) for (topic, score) in naive.items()}
self.assertEqual(self.user.topics, topics)
def test_classify(self, _s3_mock):
weights = classifier.SimpleWeights.load(classifier.s3_key_xreadlines())
self.assertEqual(
weights.classify("About 75% of New York City School students qualify "
"for free or reduced-price lunch."),
{'Healthcare': 0,
'Education': atan_norm(0.6 + 0.9)} # students, school
)
def classify(self, corpus, *topics):
"""Classify `corpus` based on number of occurrences of words and phrases, and their
weights, in the SimpleWeights dictionary.
By default, `corpus` is classified for all topics for which there are weights.
Alternatively, topics may be specified as arbitrary arguments:
SIMPLE_WEIGHTS.classify(corpus, 'healthcare', 'cooking', ...)
"""
return {topic: atan_norm(score) for (topic, score) in self.iter_topics(corpus, *topics)}
def get_topics(post_interactions, post_topics):
"""Return a User's interests scored by topic, given an iterable of the user's
PostInteractions and a catalog of PostTopics.
"""
scores = collections.defaultdict(int)
for interaction in post_interactions:
try:
catalogued = post_topics[interaction.postid]
except KeyError:
topics = {}
else:
topics = catalogued.document
for (topic, value) in topics.items():
# For now, all interactions weighted the same:
for (_interaction_type, count) in interaction.document.items():
scores[topic] += value * count
# Normalize topic scores to 1:
return {topic: atan_norm(value) for (topic, value) in scores.items()}
