def __init__(self, matrix, similar = 0):
"""
init method
Args:
matrix: the user-item matrix contain the score by user.
matrix is an instance of recsys.utils.sparse_matrix.DictMatrix.
if you make sure matrix's index i and j is continuous from 0, the compute speed while be fast.
"""
self._matrix = matrix
self._avg = {}
if similar == UserCF.USER_SIMILAR:
from recsys.similarity.user_similar import UserSimilar
self._user_similar = UserSimilar(matrix)
elif similar == UserCF.HERLOCKER_USER_SIMILAR:
from recsys.similarity.herlocker_user_similar import HerlockerUserSimilar
self._user_similar = HerlockerUserSimilar(matrix)
def setUp(self):
db = bsddb.btopen(None, cachesize = 268435456)
matrix = DictMatrix(10, 10, db)
m1, m2 = ([1, 1, 2, 3], [1, 1, 2, 3])
for i in range(1,3):
for j in range(1,5):
if i == 1:
matrix[i, j] = m1[j-1]
else:
matrix[i, j] = m2[j-1]
self.similar = HerlockerUserSimilar(matrix)
class HerlockerUserSimilarTestCase(unittest.TestCase):
def setUp(self):
db = bsddb.btopen(None, cachesize = 268435456)
matrix = DictMatrix(10, 10, db)
m1, m2 = ([1, 1, 2, 3], [1, 1, 2, 3])
for i in range(1,3):
for j in range(1,5):
if i == 1:
matrix[i, j] = m1[j-1]
else:
matrix[i, j] = m2[j-1]
self.similar = HerlockerUserSimilar(matrix)
def tearDown(self):
self.similar = None
def testCompute(self):
self.assertEqual(True, abs(self.similar.compute(1,2) - 0.08) < 0.01)
class UserCF():
"""
Use user-cf algorithm to recommend
Attributes:
_matrix: the user-item matrix contain the score by user, an instance of recsys.utils.sparse_matrix.DictMatrix.
_user_similar: instance of UserSimilar
"""
#static final value for use different similar algorithm
USER_SIMILAR = 0 #UserSimilar
HERLOCKER_USER_SIMILAR = 1 #HerlockerUserSimilar
def __init__(self, matrix, similar = 0):
"""
init method
Args:
matrix: the user-item matrix contain the score by user.
matrix is an instance of recsys.utils.sparse_matrix.DictMatrix.
if you make sure matrix's index i and j is continuous from 0, the compute speed while be fast.
"""
self._matrix = matrix
self._avg = {}
if similar == UserCF.USER_SIMILAR:
from recsys.similarity.user_similar import UserSimilar
self._user_similar = UserSimilar(matrix)
elif similar == UserCF.HERLOCKER_USER_SIMILAR:
from recsys.similarity.herlocker_user_similar import HerlockerUserSimilar
self._user_similar = HerlockerUserSimilar(matrix)
def prediction(self, user, item):
"""
get prediction of user-item pair by user-cf
prediciton(a, p) = avg_a + sum( sim(a,b) * (r(b,p) - avg_b) / sum(sim(a,b))
Args:
user: user index
item: item index
"""
# if user have rate the item
if self._matrix[user, item] != 0:
return self._matrix[user, item]
# else
topN = self._user_similar.topN(user, item, 50)
avg_u = self._userAverage(user)
count = 0.0
weights = 0.0
for i in topN:
count += (self._matrix[i[0], item] - self._userAverage(i[0])) * i[1]
weights += i[1]
if weights == 0:
return avg_u
return avg_u + count / weights
def _userAverage(self, user):
"""
compute the average rate of user
Args:
user: user index
"""
try:
return self._avg[user]
except:
avg = self._average(self._matrix[user, ...])
self._avg[user] = avg
return avg
def _average(self, vec):
"""
compute the average of vec
Args:
vec: dictionary with key-index, value-rate
"""
vec_len = max(vec.keys())+1
total = 0.0
count = 0
for i in range(vec_len):
try:
if vec[i] <> 0:
total += vec[i]
count += 1
except:
pass
if count == 0:
return 0
return total/count
