How to speed up this ddply?

Question

I have dataframe about the similarity of pair of products such as:

   Product1   Product2  similarity   
   p1          p2         0.102
   p1          p3         0.221
   p1          p4         0.333 
   .....
   p2          p1         0.102
   p2          p3         0.201
   p2          p4         0.242 

I would like to choose the top 10 most similar product for each product, such as

product.pairs<-ddply(product.pairs, "product1", transform, rank = seq_along(product1)) 
product.pairs<-subset(product.pairs, rank<11,select=c(product1,product2))

this worked when the dataset is small, but once the product number reach 30k, it is too slow...

I also tried on sqldf, to mimic the rank & partition, such as...

sql_top10<-" select a.product1,a.product2, a.similarity,count(*) as rank from productpairs a join productpairs b on a.product1=b.product1 and a.similarity>=b.similarity group by a.product1,a.simlarity"

but this even worse... Any suggestions?

Answer 1

Use data.table:

library(data.table)

dt = data.table(your_df)

# fast sort by similarity
setkey(dt, similarity)

# pick (at most) top 10 most similar ones
dt[, Product2[max(1, .N-9):.N], by = Product1]