Un'altra alternativa:
alex = function(x)
{
ux = unique(x)
mux = match(x, ux)
ans = integer(length(ux))
for(i in seq_along(x)) ans[mux[i]] = i - ans[mux[i]]
return(setNames(ans - 1L, ux))
}
alex(y)
# E F A D C B G H
#11 6 6 10 1 4 3 1
Rispetto alle altre alternative:
frank1 = function(x) tapply(1:length(x), x, diff) - 1
library(data.table)
frank2 = function(x) data.table(x)[, diff(.I) - 1, by = x]
jaehyeon = function(x) sapply(unique(x), function(X) abs(diff(which(x == X))) - 1)
library(data.table)
khashaa = function(x)
{
ux = unique(x)
setNames(length(x) - chmatch(ux, rev(x)) - chmatch(ux, x), ux)
}
khashaa_base = function(x)
{
ux = unique(x)
setNames(length(x) - match(ux, rev(x)) - match(ux, x), ux)
}
frank1(y)
# A B C D E F G H
# 6 4 1 10 11 6 3 1
frank2(y)
# x V1
#1: E 11
#2: F 6
#3: A 6
#4: D 10
#5: C 1
#6: B 4
#7: G 3
#8: H 1
jaehyeon(y)
# E F A D C B G H
#11 6 6 10 1 4 3 1
khashaa(y)
# E F A D C B G H
#11 6 6 10 1 4 3 1
khashaa_base(y)
# E F A D C B G H
#11 6 6 10 1 4 3 1
E su un punto di riferimento:
#compiled versions for all for consistency:
cmpalex = compiler::cmpfun(alex)
cmpfrank1 = compiler::cmpfun(frank1)
cmpfrank2 = compiler::cmpfun(frank2)
cmpjaehyeon = compiler::cmpfun(jaehyeon)
cmpkhashaa = compiler::cmpfun(khashaa)
cmpkhashaa_base = compiler::cmpfun(khashaa_base)
set.seed(007); xx = sample(rep(make.unique(rep_len(LETTERS, 1e3)), each = 2))
sort_by_names = function(x) x[order(names(x))]
sum(sort_by_names(alex(xx)) != frank1(xx))
#[1] 0
sum(alex(xx) != setNames(frank2(xx)[[2]], frank2(xx)[[1]]))
#[1] 0
sum(alex(xx) != jaehyeon(xx))
#[1] 0
sum(alex(xx) != khashaa(xx))
#[1] 0
sum(alex(xx) != khashaa_base(xx))
#[1] 0
microbenchmark::microbenchmark(alex(xx), cmpalex(xx),
frank1(xx), cmpfrank1(xx),
frank2(xx), cmpfrank2(xx),
jaehyeon(xx), cmpjaehyeon(xx),
khashaa(xx), cmpkhashaa(xx),
khashaa_base(xx), cmpkhashaa_base(xx), times = 20)
#Unit: microseconds
# expr min lq median uq max neval
# alex(xx) 3472.726 3620.1055 3764.005 4157.9445 5382.221 20
# cmpalex(xx) 1056.538 1074.6345 1115.177 1251.0720 2131.172 20
# frank1(xx) 19441.559 19858.8145 20356.808 21159.3035 27471.738 20
# cmpfrank1(xx) 19166.288 19566.4925 20572.222 21108.8430 22243.335 20
# frank2(xx) 12592.156 12931.6325 13337.057 14092.5725 24015.020 20
# cmpfrank2(xx) 12396.578 12861.3365 13376.904 14012.3575 14542.715 20
# jaehyeon(xx) 45313.525 46875.1900 47514.821 48728.3085 49513.578 20
# cmpjaehyeon(xx) 44899.401 46496.7365 47748.330 49561.9505 82592.347 20
# khashaa(xx) 189.314 204.1045 220.982 235.0760 259.959 20
# cmpkhashaa(xx) 190.010 201.3200 234.032 240.1225 389.415 20
# khashaa_base(xx) 295.802 315.1170 328.167 360.5320 1353.038 20
# cmpkhashaa_base(xx) 295.803 301.8930 317.901 332.8650 379.323 20
EDIT: Incluso/risolto altre alternative. La compilazione del codice byte ha migliorato solo la funzione con il ciclo esplicito; altre alternative sono state compilate solo per completezza. La soluzione intelligente di Khashaa è, inoltre, la più veloce finora.
Usa 'keyby' invece di' [ordine()] 'per ordinare direttamente da' y' qui .. – Arun