Differences

This shows you the differences between two versions of the page.

--- data_analysis:recommendation_bpr [2020/04/29 11:29] – prgram
+++ data_analysis:recommendation_bpr [2025/07/07 14:12] (current) – external edit 127.0.0.1
@@ Line 36: / Line 36: @@
 * $S \subseteq U \times I$ : implicit feedback
 * $>_u \subset I^2$ : personalied total ranking of all items
-   >  properties of total order:  $\forall i,j,k \in I$
+>  properties of total order:  $\forall i,j,k \in I$
-	  (totality) : $\text{if } i \neq j \text{, then } i >_u j \text{ or } j >_u i$
+> (totality) : $\text{if } i \neq j \text{, then } i >_u j \text{ or } j >_u i$
-	  (antisymmetry) : $\text{if } i>_uj \text{ and } j>_ui \text{, then } i = j$
+> (antisymmetry) : $\text{if } i>_uj \text{ and } j>_ui \text{, then } i = j$
-	  (transitivity) : $\text{if } i>_uj \text{ and } j>_uk \text{, then } i>_u k$
+> (transitivity) : $\text{if } i>_uj \text{ and } j>_uk \text{, then } i>_u k$
 * $I_u^+:=\{i \in I : (u,i) \in S\}$, $U_i^+:=\{u \in U : (u,i) \in S\}$
@@ Line 202: / Line 202: @@
 ![](img3.png)![](img_netflix3.png)
+{{:data_analysis:pasted:20200429-203305.png}}
+{{:data_analysis:pasted:20200429-203230.png}}
 ###### remove rating scores, repeated 5 times, $\lambda = 0.0025$, $\alpha = 0.05$
 ---
@@ Line 208: / Line 209: @@
 # END
+==== Code ====
+<code>
+#install.packages("rrecsys", force=TRUE, dependencies=TRUE)
+test_index<- function(data, na=TRUE) {
+  index_test<-c()
+  for (i in 1:dim(data)[1]) {
+    if(na) indeX_i <- which(!is.na(data[i,]))
+    else index_i<-which(data[i,]==0)
+    index_test[i]<-index_i[sample(1:length(index_i),1)]
+  }
+  test_index <- index_test
+}
+train <- function (data, index_test, na=TRUE) {
+  train_data<-data
+  for (i in 1:dim(data)[1]) {
+    train_data[i,index_test[i]]<-NA
+  }
+  train_data[is.na(train_data)]<-0
+  train<-train_data
+}
+calc_AUC <- function(data, pred, index_test, na=TRUE) {
+  AUC_u <- 0
+  for (i in 1:dim(data)[1]) {
+    i=1
+    if(na) index_j<-which(is.na(data[i,]))
+    else  index_j<-which(data[i,]==0)
+    AUC_u <- AUC_u + 1/length(index_j) * sum(pred[i,index_test[i]] > pred[i,index_j])
+  }
+  calc_AUC <- AUC_u / dim(data)[1]
+}
+############################################
+############################################
+############################################
+############################################
+############################################
+# check data
+which(apply(is.na(data2),1,sum)>1680)
+which(apply(is.na(data2),2,sum)<10)
+sum(!is.na(data2))
+############################################
+############################################
+ml100k_ori<-ml100k
+ml100k[ml100k>0]<-1
+index_test<- test_index(ml100k, na=FALSE)
+data2_train<-train(ml100k,index_test, na=FALSE)
+############################################
+############################################
+############################################
+############################################
+####BPR
+library("rrecsys")
+setwd("C:/Users/heeseoklee/Downloads/rrecsys-master/rrecsys-master/R")
+source("ALG_BPR.R")  #BPR2
+showDeltaError()
+#rrecsys(ratings, alg="BPR")
+ratings<-defineData(data2_train)
+AUC_total<-c()
+for (i in 1:10) {
+  AUC<-0
+  for( j in 1:5){
+    model_bpr<-BPR2(ratings,k=10*i)
+    pred_bpr<-predict(model_bpr)
+    AUC<-AUC+calc_AUC(ml100k,pred_bpr,index_test,na=FALSE)
+  }
+  AUC_total[i]<-AUC/5
+}
+AUC_total_bpr<-AUC_total
+setStoppingCriteria(autoConverge = FALSE,
+                    deltaErrorThreshold = 1e-05, nrLoops = NULL, minNrLoops = 10)
+showStoppingCriteria()
+############################################
+############################################
+####WRMF
+#devtools::install_github("dselivanov/reco")
+#library("reco")
+install.packages("devtools")
+####R CMD INSTALL rsparse-master/
+devtools::install_github("dselivanov/rsparse", force=TRUE)
+library(rsparse)
+data("movielens100k")
+library(Matrix)
+Matrix(movielens100k)
+ml100k<-as(movielens100k, "matrix")#[1:10,1:10]
+AUC_total<-c()
+for (i in 1:10) {
+  AUC<-0
+  for( j in 1:5){
+    fb<-"implicit"
+    #fb<-"explicit"
+    model = WRMF$new(rank = i*10, lambda = 0.0025,
+                     feedback = fb,#c("implicit", "explicit"),
+                     init_stdv = 0.01,
+                     n_threads = parallel::detectCores(),
+                     non_negative = TRUE,
+                     solver = "conjugate_gradient",#c("conjugate_gradient", "cholesky"),
+                     cg_steps = 5L,
+                     components = NULL)
+    x<-as(data2_train, "CsparseMatrix")
+    #x<-movielens100k
+    n_users_rank<-model$fit_transform(x, n_iter = 10L)
+    rank_n_items<-model$components
+    pred_mmmf<-n_users_rank %*% rank_n_items
+    AUC<-AUC+calc_AUC(ml100k,pred_mmmf,index_test,na=FALSE)
+  }
+  AUC_total[i]<-AUC/5
+}
+AUC_total_wrmf<-AUC_total
+############################################
+############################################
+####SVD-MF
+#install.packages("recommenderlab")
+library("recommenderlab")
+ml <- as(data2_train,"realRatingMatrix")
+r_svd <- Recommender(ml, method = "SVD",param=list(k=10))
+p.svd <- predict(r_svd, ml, type = "ratings")
+pred_svd<-getModel(r_svd)$svd$u %&% diag((getModel(r_svd)$svd)$d) %*% t(getModel(r_svd)$svd$v)
+AUC<-calc_AUC(ml100k,pred_svd,index_test,na=FALSE)
+r <- Recommender(data = ml, method = "ALS",
+                 parameter = list(normalize=NULL, lambda=0.0025, n_factors=10,
+                                  n_iterations=10, seed = NULL, verbose = FALSE))
+r
+recom_ratings <- predict(r, newdata = ml, type = "ratings")
+as(recom_ratings, "matrix")
+getModel(r)
+AUC_total<-c()
+for (i in 1:10) {
+  AUC<-0
+  for( j in 1:2){
+    fsvd<-funkSVD(ml, k = 10, gamma = 0.0025, lambda = 0.01,
+               min_improvement = 1e-06, min_epochs = 30, max_epochs = 50,
+               verbose = TRUE)
+    pred_svdf <- tcrossprod(fsvd$U, fsvd$V)
+    AUC<-AUC+calc_AUC(ml100k,pred_svdf,index_test,na=FALSE)
+  }
+  AUC_total[i]<-AUC/2
+}
+AUC_total_svdf<-AUC_total
+</code>