-
Notifications
You must be signed in to change notification settings - Fork 0
/
SlopeOne.py
208 lines (181 loc) · 6.04 KB
/
SlopeOne.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
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
import operator
import pickle
import math
import numpy as np
from Loader import Loader
from sklearn import cross_validation
FILE_PATH = '../train/ratings.csv'
PREDICT_PATH = '../prediction/prediction.csv'
FILE_PATH2 = '../train/ratings2.csv'
TOPN_PATH = '../prediction/list-top-n.txt'
class SlopeOne(object):
def __init__(self, loader):
self.diffs = {}
self.freqs = {}
self.data = loader.data
self.loader = loader
self.avg = None
self.train_idx = None
self.test_idx = None
self.item_data = None
self.k = 10
self.predict_data = loader.predict
def similarities(self):
for u,ratings in self.data.items():
length = len(ratings.items())
print str.format('\tProcessing user {0}/{1}', u, len(self.data.items()))
if length > 1500:
print 'Too many items'
else:
for (item, rating) in ratings.items():
self.freqs.setdefault(item, {})
self.diffs.setdefault(item, {})
for (item2, rating2) in ratings.items():
if item != item2:
self.freqs[item].setdefault(item2, 0)
self.diffs[item].setdefault(item2, 0.0)
self.freqs[item][item2] += 1
self.diffs[item][item2] += rating - rating2
def predict(self, u, i, default=True):
norm = 0
score = 0
#print str.format('Predicting User: {0} - Item {1}', u, i)
for item,rating in self.data[u].items():
if i != item and self.freqs.has_key(i) and self.freqs[i].has_key(item) and self.freqs[i][item] > 0:
norm += self.freqs[i][item]
score += (rating + self.diffs[i][item])*self.freqs[i][item]
if norm > 0:
return score/norm
elif default:
return self.avg[u]
else:
return 8
def top10_idx(self, user):
predictions = []
for item in self.data[user].keys():
score = self.predict(user, item, default=False)
predictions.append((item, score))
predictions = sorted(predictions, key=operator.itemgetter(1))
predictions.reverse()
return [i for i,s in predictions[0:10]]
def relevants(self, user):
rel = []
mean = np.mean(self.data[user].values())
std = np.std(self.data[user].values())
for i,score in self.data[user].items():
if score <= mean+std:
rel.append(i)
return rel
def test_error(self, test):
RMSE = 0.
count = 0
for n,item in enumerate(test):
if n%1000 == 0: print str.format('\tProcessing RMSE item {0}/{1}', n, len(test))
#print str.format('\tProcessing RMSE item {0}/{1}', n, len(test))
for user,score in self.item_data[item].items():
score = self.predict(user, item)
score = min(score, 10)
score = max(score, 1)
RMSE += (score - self.data[user][item])**2
count += 1
return math.sqrt(RMSE/count)
def test_topN(self, test):
top_users = set()
precision = 0.
n = 0
for n,item in enumerate(test):
if n%1000 == 0: print str.format('\tProcessing top10 item {0}/{1}', n, len(test))
for user,rating in self.item_data[item].items():
if rating >= np.mean(self.data[user].values())+np.std(self.data[user].values()) and not user in top_users:
top_users.add(user)
top10 = self.top10_idx(user)
relevants = self.relevants(user)
precision += len(np.intersect1d(top10, relevants)) / 10
n += 1
return precision / n
def predict_file(self):
self.loader.load_predict()
length = len(self.loader.predict.items())
file = open("results/prediction.csv", "wb")
for user,items in self.loader.predict.items():
print str.format('Predicting user {0}/{1}', self.loader.user_idx(user), length)
for item in items.keys():
try:
score = self.predict(self.loader.user_idx(user), self.loader.item_idx(item))
score = min(score, 10)
score = max(score, 1)
file.write(str.format('"{0}";"{1}";{2}\n', user, item, score))
except Exception:
score = np.mean(self.data[self.loader.user_idx(user)].values())
file.write(str.format('"{0}";"{1}";{2}\n', user, item, score))
def mean(self):
mean = {}
n = {}
for user,rating in self.data.items():
for i,r in rating.items():
mean.setdefault(i, 0)
n.setdefault(i, 0)
mean[i] += r
n[i] += 1
self.avg = []
for i,v in mean.items():
self.avg.append(v/n[i])
print len(self.avg)
def top10(self):
top10 = {}
length = len(self.loader.predict.items())
for user,items in self.loader.predict.items():
print str.format('Predicting top10 user {0}/{1}', self.loader.user_idx(user), length)
predictions = []
for item in self.loader.items():
i = self.loader.item_idx(item)
u = self.loader.user_idx(user)
score = self.predict(u, i)
predictions.append((item, score))
predictions = sorted(predictions, key=operator.itemgetter(1))
predictions.reverse()
top10[user] = predictions[0:10]
file = open("temp/slope_one/top10.txt", "wb")
for user,items in top10.items():
file.write(str.format("{0}\n", user))
for data in items:
file.write(str.format('\t"{0}"\n', data[0]))
if __name__ == '__main__':
loader = Loader(FILE_PATH, PREDICT_PATH)
loader.load_user_base()
slope_one = SlopeOne(loader)
loader2 = Loader(FILE_PATH, PREDICT_PATH)
loader2.load_item_base()
slope_one.item_data = loader2.data
try:
slope_one.avg = pickle.load(open('temp/slope_one/mean.p', 'rb'))
except Exception:
slope_one.mean()
pickle.dump(slope_one.avg, open('temp/slope_one/mean.p', 'wb'))
try:
print '> Loading dev matrix'
f = open('temp/slope_one/freqs.p', 'rb')
slope_one.freqs = pickle.load(f)
f.close()
f = open('temp/slope_one/diffs.p', 'rb')
slope_one.diffs = pickle.load(f)
f.close()
print '> Loaded Slop One'
except Exception, e:
print e
slope_one.similarities()
pickle.dump(slope_one.freqs, open('temp/slope_one/freqs.p', 'wb'))
pickle.dump(slope_one.diffs, open('temp/slope_one/diffs.p', 'wb'))
# print '> Testing model'
# kf = cross_validation.KFold(len(slope_one.loader.items), n_folds=5)
# RMSE = 0.
# precision = 0.
# for train_index, test_index in kf:
# slope_one.train_idx = train_index
# slope_one.test_idx = test_index
# RMSE += slope_one.test_error(test_index)
# precision += slope_one.test_topN(test_index)
# print 'RMSE = ', RMSE/5
# print 'Precison@10', precision/5
print '> Predicting File'
slope_one.predict_file()