talegari/slimrec

Sparse Linear Method to Predict Ratings and Top-N Reccomendations

slimrec 0.1.0

slimrec's Issues

[enhancement] Integrate with `recommenderlab`

Integrate with recommenderlab

[enhancement] Add functionality to check RMSE by perturbing some 1's

Consider implicit feedback dataset. Randomly(or deterministically) change a few 1's to 0's. After slim ouputs a matrix, check RMSE at those positions only.

library("slimrec")
data(ft_small)
# making 10 percent of 1s as zeros
index_ones       <- which(as.matrix(ft_small) == 1)
set.seed(1)
index            <- sample(index_ones, length(index_ones)/10)
ft_small2        <- ft_small
ft_small2[index] <- 0
# 777 1's are converted to zeroes
temp <- slim(mat = as(ft_small2, "dgCMatrix"), returnMat = TRUE, nonNegCoeff = FALSE)
ModelMetrics::rmse(temp$ratingMat[index], ft_small[index]) 
# gives me an RMSE around 0.6453588

Thanks to Kevin Hao for suggesting this.

[enhancement] Add error metrics like MPR

A source: Solving the apparent diversity-accuracy dilemma of recommender systems, by Tao Zhou et al 2010.