increase the speed of monte carlo choice simulation

[asiripanich]

The matrix approach is much faster than the sampling approach.

probabilities <- matrix(runif(1000000), 100000, 10)
probabilities_ <- probabilities - runif(100000 * 10)

ind_sapply <- sapply(1:nrow(probabilities), function(i) sample(1:10, 1, prob = probabilities[i,]))
ind_mat <- max.col(probabilities_, 'first')

table(ind_sapply)
table(ind_mat)

Unit: milliseconds
                                                                                       expr        min         lq       mean
 sapply(1:nrow(probabilities), function(i) sample(1:10, 1, prob = probabilities[i,      ])) 469.872150 619.764793 656.780946
                                                           max.col(probabilities_, "first")   1.955379   2.209599   2.308426
     median        uq         max neval cld
 655.096765 686.15894 1047.475831   100   b
   2.270803   2.36885    2.911497   100  a