/
nonnormalized_cosine_neighborhood.py
142 lines (114 loc) · 5 KB
/
nonnormalized_cosine_neighborhood.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
'''
Baseline neighborhood-based predictor from
Cremonesi, Koren, and Turrin (RecSys 2010)
'''
from predictor import Predictor
from ranker import Ranker
from evaluator import Evaluator
import sys
import numpy as np
import math
class NNCossNgbrPredictor(object):
def __init__(self, i, u):
self.item_ratings = i
self.user_ratings = u
self.__extractSimilarItems(threshold=100)
def __similarity(self, item1, item2, shrinking_factor=100.0):
ratings1 = []
ratings2 = []
for user in self.item_ratings[item1]:
if user in self.item_ratings[item2]:
ratings1.append(self.item_ratings[item1][user])
ratings2.append(self.item_ratings[item2][user])
if len(ratings1) == 0: return 0.0
cosine = np.inner(ratings1,ratings2)/(math.sqrt(np.inner(ratings1, ratings1)) * math.sqrt(np.inner(ratings2, ratings2)))
return ((1.0 * len(ratings1))/(1.0 * shrinking_factor + len(ratings1))) * cosine
def __topMatches(self, item, threshold):
scores=[(self.__similarity(item,other, shrinking_factor=100), other)
for other in self.item_ratings.keys() if other != item]
scores.sort(); scores.reverse();
return dict((y, x) for x, y in scores[:threshold])
def __extractSimilarItems(self, threshold=60):
self.similar_items = {}
c = 0
for item in self.item_ratings:
# Status updates for large datasets
c += 1
if c % 100 == 0: print "%d / %d" % (c,len(self.item_ratings))
self.similar_items[item] = self.__topMatches(item,threshold)
def get_recommendations(self, person, item_threshold=10):
user_ratings = self.user_ratings[person]
scores = {}
# Loop over items rated by this user
for (item, rating) in user_ratings.items():
# Loop over items similar to this one
try:
for (similarity,item2) in self.similar_items[item][:item_threshold]:
if item2 in user_ratings: continue
scores.setdefault(item2,0.0)
scores[item2] += similarity * rating
except:
continue
rankings = [(score,item) for item,score in scores.items()]
rankings.sort(); rankings.reverse()
return rankings
def get_ratings(self, user, selected_items):
'''
applies NNCosNgbr, but instead of searching items in similar_items
within a threshold, searches for all selected_items and generates
scores for all of them.
'''
user_ratings = []
try:
user_ratings = self.user_ratings[user]
except:
return []
scores = {}
# Loop over items rated by this user
for (item, rating) in user_ratings.items():
for item2 in selected_items:
scores.setdefault(item2,0)
if item2 in self.similar_items[item]:
scores[item2] += self.similar_items[item][item2] * rating
rankings = [(score,item) for item,score in scores.items()]
rankings.sort(); rankings.reverse()
return rankings
if __name__=="__main__":
'''
sys.argv[1] => training data
sys.argv[2] => test data
sys.argv[3] => data separator
'''
training = Predictor(sys.argv[1], sys.argv[3])
training_users, training_items = training.store_data_relations() #~100MB
recommender = NNCossNgbrPredictor(training_items, training_users)
N = 10
ranker = Ranker(N)
testing = Predictor(sys.argv[2], sys.argv[3])
test_users, test_items = testing.store_data_relations()
ev = Evaluator(test_users, N)
#TODO clean this interface!
item_ids = list(set(training_items.keys() + test_items.keys())) #all unique items in the dataset
hits = 0
div_metric1 = []
div_metric2 = []
recommended_ratings = []
for u in test_users.keys():
for i in test_users[u].keys():
user_items = []
if u in training_users:
user_items = training_users[u].keys()
if u in test_users:
user_items += test_users[u].keys()
items_for_cremonesi_validation = testing.choose_some_items(item_ids, user_items, i, 40)
ratings = recommender.get_ratings(u, items_for_cremonesi_validation)
recommendations = ranker.topRatings(ratings)
#recommendations = ranker.maximizeKGreatItems(1, ratings, training_items)
recommended_ratings += ev.totalOfRatings(u, recommendations)
hits += ev.hadAHit(recommendations, i)
div_metric1.append(ev.simpleDiversity(recommendations, training_items))
div_metric2.append(ev.diversityEILD(recommendations, training_items))
test_size = 3191.0
print 'rec', hits/test_size, 'prec', hits/(test_size * N)
print 'sim simple', sum(div_metric1)/len(div_metric1)
print 'div vargas', sum(div_metric2)/len(div_metric2)