def setUp(self):
self.k = 40
self.recommender = FMRecommender(p=3, k=self.k)
self.recommender.initialize()
logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s')
logging.root.setLevel(level=logging.INFO)
logger.info('running %s' % ' '.join(sys.argv))
if len(sys.argv) < 2:
print(globals()['__doc__'] % locals())
sys.exit(1)
logging.info('converting data into FluRS input object')
data = fetch_movielens(data_home=sys.argv[1], size='1m')
logging.info('initialize recommendation model and evaluation module')
rec = FMRecommender(
p=sum(data.contexts.values()), # number of dimensions of input vector
k=40,
l2_reg_w0=2.,
l2_reg_w=8.,
l2_reg_V=16.,
learn_rate=.004)
rec.initialize()
evaluator = Evaluator(rec, data.can_repeat)
n_batch_train = int(data.n_sample *
0.2) # 20% for pre-training to avoid cold-start
n_batch_test = int(data.n_sample *
0.1) # 10% for evaluation of pre-training
batch_tail = n_batch_train + n_batch_test
# pre-train
# 20% for batch training | 10% for batch evaluate
# after the batch training, 10% samples are used for incremental updating
class FMRecommenderTestCase(TestCase):
def setUp(self):
self.k = 40
self.recommender = FMRecommender(p=3, k=self.k)
self.recommender.initialize()
def test_register_user(self):
self.recommender.register(User(0))
self.assertEqual(self.recommender.n_user, 1)
def test_register_item(self):
self.recommender.register(Item(0))
self.assertEqual(self.recommender.n_item, 1)
def test_update(self):
self.recommender.register(User(0))
self.recommender.register(Item(0))
self.recommender.update(Event(User(0), Item(0), 1))
self.assertEqual(self.recommender.n_user, 1)
self.assertEqual(self.recommender.n_item, 1)
def test_score(self):
self.recommender.register(User(0))
self.recommender.register(Item(0))
self.recommender.update(Event(User(0), Item(0), 1))
score = self.recommender.score(User(0), np.array([0]), np.array([0]))
self.assertTrue(score >= 0.)
with open('h.pickle', 'wb') as f:
pickle.dump(h, f)
with open('data.pickle', 'wb') as f:
pickle.dump(data, f)
if os.path.exists('model.pickle'):
print('Using pickled model')
with open('model.pickle', 'rb') as f:
rec = pickle.load(f)
else:
print('Training...')
rec = FMRecommender(
p=sum(data.contexts.values()), # number of dimensions of input vector
k=60,
l2_reg_w0=2.,
l2_reg_w=8.,
l2_reg_V=16.,
learn_rate=.004)
rec.initialize()
n_batch_train = int(data.n_sample * 0.9)
evaluator = Evaluator(rec, data.can_repeat)
evaluator.debug = True
evaluator.fit(data.samples[:n_batch_train],
data.samples[n_batch_train:],
n_epoch=4)
print('Done training!')
with open('model.pickle', 'wb') as f:
pickle.dump(rec, f)