Optimization of an R loop taking 18 hours to run - r

Question

I've got an R code that works and does what I want but It takes a huge time to run. Here is an explanation of what the code does and the code itself.
I've got a vector of 200000 line containing street adresses (String) : data.
Example :
> data[150000,]
address
"15 rue andre lalande residence marguerite yourcenar 91000 evry france"
And I have a matrix of 131x2 string elements which are 5grams (part of word) and the ids of the bags of NGrams (example of a 5Grams bag : ["stack", "tacko", "ackov", "ckover", ",overf", ... ] ) : list_ngrams
Example of list_ngrams :
idSac ngram
1 4 stree
2 4 tree_
3 4 _stre
4 4 treet
5 5 avenu
6 5 _aven
7 5 venue
8 5 enue_
I have also a 200000x31 numerical matrix initialized with 0 : idv_x_bags
In total I have 131 5-grams and 31 bags of 5-grams.
I want to loop the string addresses and check whether it contains one of the n-grams in my list or not. If it does, I put one in the corresponding column which represents the id of the bag that contains the 5-gram.
Example :
In this address : "15 rue andre lalande residence marguerite yourcenar 91000 evry france". The word "residence" exists in the bag ["resid","eside","dence",...] which the id is 5. So I'm gonna put 1 in the column called 5. Therefore the corresponding line "idv_x_bags" matrix will look like the following :
> idv_x_sacs[150000,]
4 5 6 8 10 12 13 15 17 18 22 26 29 34 35 36 42 43 45 46 47 48 52 55 81 82 108 114 119 122 123
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Here is the code that does :
idv_x_sacs <- matrix(rep(0,nrow(data)*31),nrow=nrow(data),ncol=31)
colnames(idv_x_sacs) <- as.vector(sqldf("select distinct idSac from list_ngrams order by idSac"))$idSac
for(i in 1:nrow(idv_x_bags))
{
for(ngram in list_ngrams$ngram)
{
if(grepl(ngram,data[i,])==TRUE)
{
idSac <- sqldf(sprintf("select idSac from list_ngramswhere ngram='%s'",ngram))[[1]]
idv_x_bags[i,as.character(idSac)] <- 1
}
}
}
The code does perfectly what I aim to do, but it takes about 18 hours which is huge. I tried to recode it with c++ using Rcpp library but I encountered many problems. I'm tried to recode it using apply, but I couldn't do it.
Here is what I did :
apply(cbind(data,1:nrow(data),1,function(x){
apply(list_ngrams,1,function(y){
if(grepl(y[2],x[1])==TRUE){idv_x_bags[x[2],str_trim(as.character(y[1]))]<-1}
})
})
I need some help with coding my loop using apply or some other method that run faster that the current one. Thank you very much.

Answer 1

Check this one and run the simple example step by step to see how it works.
My N-Grams don't make much sense, but it will work with actual N_Grams as well.
library(dplyr)
library(reshape2)
# your example dataset
dt_sen = data.frame(sen = c("this is a good thing", "this is bad"), stringsAsFactors = F)
dt_ngr = data.frame(id_ngr = c(2,2,2,3,3,3),
ngr = c("th","go","tt","drf","ytu","bad"), stringsAsFactors = F)
# sentence dataset
dt_sen
sen
1 this is a good thing
2 this is bad
#ngrams dataset
dt_ngr
id_ngr ngr
1 2 th
2 2 go
3 2 tt
4 3 drf
5 3 ytu
6 3 bad
# create table of matches
expand.grid(unique(dt_sen$sen), unique(dt_ngr$id_ngr)) %>%
data.frame() %>%
rename(sen = Var1,
id_ngr = Var2) %>%
left_join(dt_ngr, by = "id_ngr") %>%
group_by(sen, id_ngr,ngr) %>%
do(data.frame(match = grepl(.$ngr,.$sen))) %>%
group_by(sen,id_ngr) %>%
summarise(sum_success = sum(match)) %>%
mutate(match = ifelse(sum_success > 0,1,0)) -> dt_full
dt_full
Source: local data frame [4 x 4]
Groups: sen
sen id_ngr sum_success match
1 this is a good thing 2 2 1
2 this is a good thing 3 0 0
3 this is bad 2 1 1
4 this is bad 3 1 1
# reshape table
dt_full %>% dcast(., sen~id_ngr, value.var = "match")
sen 2 3
1 this is a good thing 1 0
2 this is bad 1 1