acceleration for neighborhood-based methods

algorithm/rating/ItemKNN.py

@@ -60,7 +60,7 @@ def computeCorr(self):
                 if i1 <> i2:
                     if self.itemSim.contains(i1,i2):
                         continue
-                    sim = qmath.similarity(self.dao.col(i1),self.dao.col(i2),self.sim)
+                    sim = qmath.similarity(self.dao.sCol(i1),self.dao.sCol(i2),self.sim)
                     self.itemSim.set(i1,i2,sim)
             print 'item '+i1+' finished.'
         print 'The item correlation has been figured out.'

algorithm/rating/UserKNN.py

@@ -60,7 +60,7 @@ def computeCorr(self):
                 if u1 <> u2:
                     if self.userSim.contains(u1,u2):
                         continue
-                    sim = qmath.similarity(self.dao.row(u1),self.dao.row(u2),self.sim)
+                    sim = qmath.similarity(self.dao.sRow(u1),self.dao.sRow(u2),self.sim)
                     self.userSim.set(u1,u2,sim)
             print 'user '+u1+' finished.'
         print 'The user correlation has been figured out.'

baseclass/IterativeRecommender.py

@@ -53,6 +53,10 @@ def updateLearningRate(self,iter):
     def predict(self,u,i):
         if self.dao.containsUser(u) and self.dao.containsItem(i):
             return self.P[self.dao.getUserId(u)].dot(self.Q[self.dao.getItemId(i)])
+        elif self.dao.containsUser(u) and not self.dao.containsItem(i):
+            return self.dao.userMeans[u]
+        elif not self.dao.containsUser(u) and self.dao.containsItem(i):
+            return self.dao.itemMeans[i]
         else:
             return self.dao.globalMean
 

config/CoFactor.conf

@@ -1,11 +1,11 @@
-ratings=../dataset/FilmTrust/trainset.txt
+ratings=../dataset/FilmTrust/trainset2.txt
 ratings.setup=-columns 0 1 2
 recommender=CoFactor
-evaluation.setup=-testSet ../dataset/FilmTrust/testset.txt
+evaluation.setup=-testSet ../dataset/FilmTrust/testset2.txt
 item.ranking=on -topN 50
 num.factors=5
-num.max.iter=5
+num.max.iter=30
 learnRate=-init 0.05 -max 1
-reg.lambda=-u 0.03 -i 0.03 -b 0.03 -s 0.1
-CoFactor=-k 3 -gamma 0.03 -filter 10
+reg.lambda=-u 0.01 -i 0.01 -b 0.01 -s 0.1
+CoFactor=-k 5 -gamma 0.03 -filter 10
 output.setup=on -dir ../results/

config/PMF.conf

@@ -1,10 +1,10 @@
-ratings=../dataset/FilmTrust/trainset.txt
+ratings=../dataset/FilmTrust/trainset2.txt
 ratings.setup=-columns 0 1 2
 recommender=PMF
-evaluation.setup=-testSet ../dataset/FilmTrust/testset.txt
-item.ranking=on -topN 10
+evaluation.setup=-testSet ../dataset/FilmTrust/testset2.txt
+item.ranking=on -topN 50
 num.factors=5
-num.max.iter=5
-learnRate=-init 0.1 -max 1
-reg.lambda=-u 0.03 -i 0.03 -b 0.1 -s 0.1
+num.max.iter=30
+learnRate=-init 0.05 -max 1
+reg.lambda=-u 0.01 -i 0.01 -b 0.1 -s 0.1
 output.setup=on -dir ../results/

config/SVD.conf

@@ -1,10 +1,10 @@
-ratings=../dataset/FilmTrust/trainset.txt
+ratings=../dataset/FilmTrust/trainset2.txt
 ratings.setup=-columns 0 1 2
 recommender=SVD
-evaluation.setup=-testSet ../dataset/FilmTrust/testset.txt
+evaluation.setup=-testSet ../dataset/FilmTrust/testset2.txt
 item.ranking=on -topN 10
 num.factors=5
-num.max.iter=10
-learnRate=-init 0.005 -max 1
-reg.lambda=-u 0.05 -i 0.05 -b 0.05 -s 0.1
+num.max.iter=30
+learnRate=-init 0.01 -max 1
+reg.lambda=-u 0.02 -i 0.02 -b 0.02 -s 0.1
 output.setup=on -dir ../results/

data/rating.py

@@ -169,6 +169,12 @@ def row(self,u):
     def col(self,c):
         return self.trainingMatrix.col(self.getItemId(c))
 
+    def sRow(self,u):
+        return self.trainingMatrix.sRow(self.getUserId(u))
+
+    def sCol(self,c):
+        return self.trainingMatrix.sCol(self.getItemId(c))
+
     def rating(self,u,c):
         return self.trainingMatrix.elem(self.getUserId(u),self.getItemId(c))
 

structure/new_sparseMatrix.py

@@ -1,4 +1,5 @@
 import numpy as np
+#class Triple(object):
 
 
 class SparseMatrix():
@@ -16,6 +17,20 @@ def __init__(self,triple):
         self.elemNum = len(triple)
         self.size = (len(self.matrix_User),len(self.matrix_Item))
 
+    def sRow(self,r):
+        if not self.matrix_User.has_key(r):
+            return {}
+        else:
+            return self.matrix_User[r]
+
+    def sCol(self,c):
+        if not self.matrix_Item.has_key(c):
+            return {}
+        else:
+            return self.matrix_Item[c]
+
+
+
     def row(self,r):
         if not self.matrix_User.has_key(r):
             return np.zeros((1,self.size[1]))

tool/qmath.py

@@ -17,6 +17,22 @@ def common(x1,x2):
     new_x2 = x2[common]
     return new_x1,new_x2
 
+def cosine_sp(x1,x2):
+    'x1,x2 are dicts,this version is for sparse representation'
+    total = 0
+    denom1 = 0
+    denom2 =0
+    for k in x1:
+        if x2.has_key(k):
+            total+=x1[k]*x2[k]
+            denom1+=x1[k]**2
+            denom2+=x2[k]**2
+    try:
+        return (total + 0.0) / (sqrt(denom1) * sqrt(denom2))
+    except ZeroDivisionError:
+        return 0
+
+
 def cosine(x1,x2):
     #find common ratings
     new_x1, new_x2 = common(x1,x2)
@@ -67,7 +83,7 @@ def similarity(x1,x2,sim):
     if sim == 'euclidean':
         return euclidean(x1,x2)
     else:
-        return cosine(x1, x2)
+        return cosine_sp(x1, x2)
 
 
 def normalize(vec,maxVal,minVal):