def tags_similar_to_vector(self, vec):
"""
Take in a category vector, and returns a weighted vector of
associated tags in the study. You can run the `top_items()`
method of this vector to get the most associated tags.
"""
return divisi2.dot(self.get_tag_matrix().normalize_rows(offset=1.0), vec)
def docs_similar_to_vector(self, vec, study='all'):
"""
Take in a category vector, and returns a weighted vector of
associated documents in the study. You can run the `top_items()`
method of this vector to get the most associated documents.
"""
return divisi2.dot(self.get_doc_matrix(study).normalize_rows(offset=1.0), vec)
def terms_similar_to_vector(self, vec):
"""
Take in a category vector, and returns a weighted vector of
associated terms. You can run the `top_items()` method of this vector
to get the most associated terms.
"""
return divisi2.dot(self.assoc.left, vec)
def learn_iter(mat):
print "Starting learning process..."
for i in range(NUM_LEARN_ITER):
user_mat, axis_weights, movie_mat = learn(mat)
# Reconstruct the learning matrix here.
for j in range(NUM_USERS):
for k in range(NUM_MOVIES):
mat[j, k] = divisi2.dot(user_mat[i,:], movie_mat[j,:])
print "Learning process complete."
start_time = time.time()
predictions = divisi2.reconstruct(user_mat, axis_weights, movie_mat)
print "Matrix reconstruction (elapsed time: %f s)." % (time.time() - start_time)
return predictions
def domain_terms_similar_to_vector(self, vec):
"""
Take in a category vector, and returns a weighted vector of
associated terms, but leave out ones that only appear in common
sense background knowledge.
You can run the `top_items()` method of this vector
to get the most associated terms.
"""
# FIXME: this way of finding domain concepts is such a hack.
mask = np.zeros((len(self.priority),), 'b')
for i, item in enumerate(self.priority.items):
if (self.priority.priority.has_key(i) and
self.priority.priority[i] < 1e6):
mask[i] = True
return divisi2.multiply(divisi2.dot(self.assoc.left.normalize_rows(offset=1.0), vec), mask)
def vector_from_terms(self, terms):
"""
Get a category vector representing the given set of weighted terms,
expressed as (term, weight) tuples. This will apply TF-IDF weighting.
"""
total_weight = 0.0
for _, weight in terms:
total_weight += abs(weight)
vec = divisi2.DenseVector(
np.zeros((len(self.priority),)),
labels=self.priority
)
for term, weight in terms:
if term in self.priority:
index = self.priority.index(term)
tfidf_weight = weight * self.get_term_idf(term) * self.database.normalized_relevance(term)
vec[index] = tfidf_weight / total_weight
category = divisi2.dot(vec, self.assoc.left)
return category
def _cosine(self, v1, v2):
return float(divisi2.dot(v1,v2) / (norm(v1) * norm(v2)))
def _cosine(self, v1, v2):
return float(divisi2.dot(v1, v2) / (norm(v1) * norm(v2)))
