def __init__(self, model_prefix='model', riter=10):
self.olda = OLDA()
self.olda.load_model(model_prefix)
self.examples = self.olda.resource_pool.examples
self.shuffle_iteration = riter
self.user_topic_vectors = self.report_theta()
self.shufflled_user_topic_vectors = []
self.user_items = []
self.blocks = self.olda.make_block_list()
# get shuffuled vectors
for i in range(riter):
self.shufflled_user_topic_vectors.append(ShuffleVectors(self.user_topic_vectors))
class Recommend:
def __init__(self, model_prefix='model', riter=10):
self.olda = OLDA()
self.olda.load_model(model_prefix)
self.examples = self.olda.resource_pool.examples
self.shuffle_iteration = riter
self.user_topic_vectors = self.report_theta()
self.shufflled_user_topic_vectors = []
self.user_items = []
self.blocks = self.olda.make_block_list()
# get shuffuled vectors
for i in range(riter):
self.shufflled_user_topic_vectors.append(ShuffleVectors(self.user_topic_vectors))
def report_theta(self):
theta = []
for m, line in enumerate(open(self.olda.model_prefix+'.theta', 'r')): # iterate number of traning examples
vals = line[:].rstrip().split(" ")
theta.append(map(float, vals))
return theta
def innerproduct(self, vec_one, vec_two):
ip = 0
for i in range(len(vec_one)):
if vec_one[i] in vec_two:
ip += 1
return ip
def get_recommend_items(self, line):
items = {}
target_uid_str, words_str = line[:].rstrip().split("\t")
target_uid = int(target_uid_str)
uvecter = self.olda.inference(line, 1000) # online learning
wvecter = self.olda.new_resource_pool.examples[0] # need rewrite...
sim_users = self.__get_similar_users(uvecter) # get the users who have the same interest as the input user
for uid in sim_users:
titems = self.__get_items(uid)
if self.innerproduct(wvecter, titems) < 1:
continue
for item in titems:
if not items.has_key(item):
items[item] = 0
items[item] += 1
rt_items = []
#item_list = items.keys()
#print "\titem_list:",item_list
i = 0
for itm, count in items.items():
if (itm in wvecter) or (itm in self.__get_items(target_uid)):
continue
if i > 20:
break
if not self.blocks.has_key(itm):
rt_items.append(itm)
i += 1
return rt_items
def __get_similar_users(self, uvecter):
susers = []
for i in range(self.shuffle_iteration):
tlist = self.shufflled_user_topic_vectors[i].get_similar_vectors(uvecter)
susers = susers + tlist
return list(set(susers)) # uniq
def __get_items(self, user_id):
return self.examples[user_id].features
