def test_RandomRecommender(self):
backend = [config_global.SAVE_RANDOM_RECOMMENDER]
self.saveJson(backend)
fb = Random_Recommender()
additional_filter_1 = {'domainid': 'domain1'}
fb.train(userid, additional_filter_1) # train for the specific user and the filter
resultSet_1 = fb.get_recommendation(userid, additional_filter_1, N=N, remove=False)
def setUp(self):
self.redis_con = redis.Redis("localhost")
self.redis_con.flushall()
userid = 123
N = 3
fb = Random_Recommender( )
additional_filter_1 = { 'domainid' : 'domain1' }
self.insertRecommendables(additional_filter_1, 0, 10)
fb.train( userid, additional_filter_1 ) # train for the specific user and the filter
def test_Get_Recommendation_with_multiple_constraints(self):
userid = 123
N = 3
fb = Random_Recommender( )
additional_filter_1 = { 'domainid' : 'domain1', 'channelid' : 'channel1' }
self.insertRecommendables(additional_filter_1, 0, 10)
fb.train( userid, additional_filter_1 ) # train for the specific user and the filter
resultSet_1 = fb.get_recommendation(userid, additional_filter_1, N=N, remove = False )
if self.debug: print resultSet_1
self.assertEqual(len(resultSet_1), N, 'the resulting recommendation have the wrong number')
def test_get_recommendable_item(self):
fb = Random_Recommender()
# inserting two items for domain1
fb.set_recommendables( 1, { 'domainid' : 'domain1' } )
fb.set_recommendables( 2, { 'domainid' : 'domain1' } )
g = fb.get_recommendable_item( { 'domainid' : 'domain1' } )
self.assertIn(int(g), (1,2), "fetched wrong recommendable item")
# for this domain there is no way of recommending anything, because we don't have any items
g = fb.get_recommendable_item( { 'domainid' : 'domain2' } )
self.assertIsNone(g, "fetched wrong recommendable item")
def test_Get_Recommendation_Ranked(self):
userid = 123
N = 3
fb = Random_Recommender( )
additional_filter_1 = { 'domainid' : 'domain1' }
self.insertRecommendables(additional_filter_1, 0, 10)
additional_filter_2 = { 'domainid' : 'domain2' }
self.insertRecommendables(additional_filter_2, 20, 30)
self.insertRecommendables({}, 20, 30)
fb.train( userid, additional_filter_1 ) # train for the specific user and the filter
resultSet_1 = fb.get_recommendation( userid, additional_filter_1, N=N, remove = False, ranked = True )
if self.debug: print resultSet_1
self.assertEqual(len(resultSet_1), N, 'the resulting recommendation have the wrong number')
def testSimpleMergeList(self):
fb = Random_Recommender( )
additional_filter_1 = { 'domainid' : 'domain1' }
userid = 123
N = 3
resultSet_1 = fb.get_recommendation( userid, additional_filter_1, N=N, remove = False, ranked = True )
if self.debug: print resultSet_1
resultSet_2 = fb.get_recommendation( userid, additional_filter_1, N=N, remove = False, ranked = True )
if self.debug: print resultSet_2
slm = SimpleListMerge()
slm.add('a', resultSet_1)
slm.add('b', resultSet_2)
unchanged_list = slm.merge_naive( {'a':0.5, 'b' : 0.5} )
self.assertEqual(unchanged_list, resultSet_1, "merging two same lists results the same list")
def test_compute_key(self):
additional_filter = { }
fb = Random_Recommender()
full_key = fb.compute_key( additional_filter )
self.assertEqual(fb.itemList , full_key, "the key constraints are wrong")
a = 'domainid'
b = 'domain2'
additional_filter = { a : b }
fb = Random_Recommender()
full_key = fb.compute_key( additional_filter )
self.assertEqual(fb.itemList + ':' + a + ':' + b, full_key, "the key constraints are wrong")
a = 'domainid'
b = 'domain2'
c = 'categoryid'
d = 'category1'
additional_filter = { a : b, c : d }
fb = Random_Recommender()
full_key = fb.compute_key( additional_filter )
self.assertEqual(fb.itemList + ':' + a + ':' + b + ':' + c + ':' + d, full_key, "the key constraints are wrong")
def test_Get_Recommendation_with_constraint(self):
userid = 123
N = 3
fb = Random_Recommender( )
additional_filter_1 = { 'domainid' : 'domain1' }
self.insertRecommendables(additional_filter_1, 0, 10)
additional_filter_2 = { 'domainid' : 'domain2' }
self.insertRecommendables(additional_filter_2, 20, 30)
self.insertRecommendables({}, 20, 30)
fb.train( userid, additional_filter_1 ) # train for the specific user and the filter
resultSet_1 = fb.get_recommendation( userid, additional_filter_1, N=N, remove = False )
if self.debug: print resultSet_1
self.assertEqual(len(resultSet_1), N, 'the resulting recommendation have the wrong number')
resultSet_2 = fb.get_recommendation( userid, additional_filter_2, N=N, remove = False )
self.assertEqual(len(resultSet_2), 0, 'the resulting recommendation have the wrong number')
resultSet_new = fb.get_recommendation( userid, {}, N=N, remove = True )
self.assertEqual(len(resultSet_new), 0, 'the resulting recommendation have the wrong number')
fb.train( userid, {} ) # train for the specific user but with no constraint, meaning we can recommend everything
resultSet_new = fb.get_recommendation( userid, {}, N=N, remove = True, ranked=True )
if self.debug: print resultSet_new
self.assertEqual(len(resultSet_new), N, 'the resulting recommendation have the wrong number')
resultSet_new_1 = fb.get_recommendation( userid, {}, N=N, remove = True, ranked=True )
if self.debug: print resultSet_new_1
self.assertEqual(len(resultSet_new_1), N, 'the resulting recommendation have the wrong number')
self.assertNotEqual(resultSet_new_1,resultSet_new, 'items were not removed from the recommendation Set')
def test_Get_Recommendation_userid_disctinction(self):
N = 3
fb = Random_Recommender( )
additional_filter_1 = { 'domainid' : 'domain1' }
userid1 = 123
self.insertRecommendables(additional_filter_1, 0, 10)
fb.train( userid1, additional_filter_1 ) # train for the specific user and the filter
userid2 = 456
self.insertRecommendables(additional_filter_1, 0, 10)
fb.train( userid2, additional_filter_1 ) # train for the specific user and the filter
resultSet_1 = fb.get_recommendation( userid1, additional_filter_1, N=N, remove = False )
if self.debug: print resultSet_1
self.assertEqual(len(resultSet_1), N, 'the resulting recommendation have the wrong number')
resultSet_2 = fb.get_recommendation( userid2, additional_filter_1, N=N, remove = False )
if self.debug: print resultSet_2
self.assertEqual(len(resultSet_2), N, 'the resulting recommendation have the wrong number')
def insertRecommendables(self, additional_filter, N1, N2): fb = Random_Recommender() for item_id in xrange(N1,N2): fb.set_recommendables( itemid = item_id, additional_filter = additional_filter )
def __init__(self, json_string, async=False, api='contest', backends=[] ):
'''
Constructor
'''
self.backends = backends
fullParsedDataModel = FullContestMessageParser()
fullParsedDataModel.parse(json_string)
domain_id = fullParsedDataModel.domain_id
user_id = fullParsedDataModel.user_id
# now compile the constraints, which are important for an onsite Recommender: do not recommend item from another domain
constraints = {'domainid': domain_id}
random_recommender = Random_Recommender()
N = 4
# TODO initialize a new training session if necessary
random_recommender.train(user_id, constraints)
self.resultSet = random_recommender.get_recommendation(user_id, constraints, N=N, remove=True)
def recommend(self):
""" lets hope there are recommendation ready for this user/constraint """
return self.resultSet
if not async: # save the data instantly
if api == 'contest':
fullParsedDataModel = FullContestMessageParser()
fullParsedDataModel.parse(message)
fullParsedDataModel.save()
item_id = fullParsedDataModel.item_id
if config_global.SAVE_RAW_JSON in backends:
raw = rawJsonModel(message, mode='redis')
raw.save()
if config_global.SAVE_RANDOM_RECOMMENDER in backends:
fb = Random_Recommender()
domain_id = fullParsedDataModel.domain_id
## todo the recommender has to decide on its own what to save and therefore save constraints, even though the constrain management should be centralized
#constraints = {'domainid': domain_id}
fb.set_recommendables(item_id, constraints)
if api == 'orp':
# todo throw not implemented error
pass
elif api == 'id_list': ## this for debugging purposes
userid = message['userid']
itemid = message['itemid']
timestamp = message['timestamp']
domainid = message['domainid']