R: Meilleure comparaison match

Question

J'ai un long fichier cadre comme ce qui suit:

df <- structure(list(Date =c("2011-01", "2011-08", "2012-03", "2011-01", "2011-08", "2011-01", "2011-08", "2011-01", "2011-08", 
        "2011-01", "2011-08", "2012-03", "2011-01", "2011-08", "2011-01", "2011-08", "2011-01", "2011-08", 
        "2011-01", "2011-08", "2012-03", "2011-01", "2011-08", "2011-01", "2011-08", "2011-01", "2011-08"), 
    Part=c("A", "A", "A", "A", "A", "A", "A", "A", "A", 
      "B", "B", "B", "B", "B", "B", "B", "B", "B", 
      "C", "C", "C", "C", "C", "C", "C", "C", "C"), 
    method=c("Type1","Type1","Type1","Type2","Type2","Type3","Type3","Type4","Type4", 
       "Type1","Type1","Type1","Type2","Type2","Type3","Type3","Type4","Type4", 
       "Type1","Type1","Type1","Type2","Type2","Type3","Type3","Type4","Type4"), 
    value= c(4L, 46L, 43L, 9L, 8L, 46L, 63L, 84L, 2L, 5L, 78L, 2L, 89L, 2L, 6L, 62L, 25L, 46L, 3L, 4L, 7L, 24L, 13L, 21L, 19L, 8L, 3L)), 
    class= "data.frame", row.names=c(NA, -27L)) 

Je voudrais créer une autre colonne appelée BestMethod. La variable devrait être une liste de méthodes correspondant à la valeur la plus proche du type 3 par partie et par date. Par exemple, en 2011-01 pour la partie A, le type 1,2,3 a été appliqué et le type 1 était le plus proche du type 3. Sous BestMethod, j'aurais le type1. Sinon, si les 3 types n'étaient pas appliqués, je mettrais NA.

(Dans Excel, il peut ressembler à ceci:

=INDEX(C2:F2, MATCH(MIN(ABS(C2:F2-B2)), ABS(C2:F2-B2),0)) 

alors ceci:

=IF(B2="", "NA", INDEX($C$1:$F$1,1,(MATCH(H2,C2:F2,0))))) 

Je voudrais créer une autre colonne appelée FinalMethod Je voudrais avoir le type le plus indiqué. Par exemple, pour 2011-01, 2011-02 pour la partie A, le type 1 correspondait mieux, mais en 2011-03, le type 2 était le meilleur. tch. Dans ce cas, je voudrais que le type 1 soit le FinalMethod pour toutes les dates de cette partie.

J'ai essayé les éléments suivants:

which(abs(x-your.number)==min(abs(x-your.number))) 

mais je suis everying mal à appeler les valeurs de données correctes et l'exécuter à travers chaque ligne.

Merci.

sortie souhaitée:

df <- structure(list(Date =c("2011-01", "2011-08", "2012-03", "2011-01", "2011-08", "2011-01", "2011-08", "2011-01", "2011-08", 
        "2011-01", "2011-08", "2012-03", "2011-01", "2011-08", "2011-01", "2011-08", "2011-01", "2011-08", 
        "2011-01", "2011-08", "2012-03", "2011-01", "2011-08", "2011-01", "2011-08", "2011-01", "2011-08"), 
    Part=c("A", "A", "A", "A", "A", "A", "A", "A", "A", 
      "B", "B", "B", "B", "B", "B", "B", "B", "B", 
      "C", "C", "C", "C", "C", "C", "C", "C", "C"), 
    method=c("Type1","Type1","Type1","Type2","Type2","Type3","Type3","Type4","Type4", 
       "Type1","Type1","Type1","Type2","Type2","Type3","Type3","Type4","Type4", 
       "Type1","Type1","Type1","Type2","Type2","Type3","Type3","Type4","Type4"), 
    value= c(4L, 46L, 43L, 9L, 8L, 46L, 63L, 84L, 2L, 5L, 78L, 2L, 89L, 2L, 6L, 62L, 25L, 46L, 3L, 4L, 7L, 24L, 13L, 21L, 19L, 8L, 3L), 
    BestModel=c("Type2", "Type1", "NA", "Type2", "Type1", "Type2", "Type1", "Type2", "Type1", 
       "Type1", "Type1Type4", "NA", "Type1", "Type1Type4", "Type1", "Type1Type4","Type1", "Type1Type4", 
       "Type2", "Type2", "NA", "Type2", "Type2", "Type2", "Type2", "Type2", "Type2"), 
    FinalModel= c("Type1Type2", "Type1Type2","Type1Type2", "Type1Type2","Type1Type2", "Type1Type2","Type1Type2","Type1Type2","Type1Type2", 
        "Type1", "Type1", "Type1", "Type1", "Type1", "Type1","Type1", "Type1", "Type1", 
        "Type2", "Type2","Type2", "Type2", "Type2", "Type2","Type2", "Type2", "Type2")), 
    class= "data.frame", row.names=c(NA, -27L)) 

Answer 1

Une solution pas très élégante en utilisant dplyr + tidyr, mais fonctionne:

library(dplyr) 
library(tidyr) 

temp = df %>% 
    group_by(Part, Date) %>% 
    mutate(value.x = ifelse(method == "Type3", value, NA)) %>% 
    fill(value.x, .direction = "up") %>% 
    fill(value.x) %>% 
    mutate(difference = abs(value.x - value)) %>% 
    filter(method != "Type3") %>% 
    filter(difference == min(difference)) 

BestMethod = temp %>% 
    summarize(BestMethod = paste(method, collapse = " ")) 

FinalMethod = temp %>% 
    group_by(Part, method) %>% 
    summarize(count = n()) %>% 
    filter(count == max(count)) %>% 
    rename(FinalMethod = method) 

df %>% 
    full_join(BestMethod) %>% 
    full_join(FinalMethod) %>% 
    select(-count) %>% 
    arrange(Part, Date) 

Résultat:

 Date Part method value BestMethod FinalMethod 
1 2011-01 A Type1  4  Type2  Type1 
2 2011-01 A Type1  4  Type2  Type2 
3 2011-01 A Type2  9  Type2  Type1 
4 2011-01 A Type2  9  Type2  Type2 
5 2011-01 A Type3 46  Type2  Type1 
6 2011-01 A Type3 46  Type2  Type2 
7 2011-01 A Type4 84  Type2  Type1 
8 2011-01 A Type4 84  Type2  Type2 
9 2011-08 A Type1 46  Type1  Type1 
10 2011-08 A Type1 46  Type1  Type2 
11 2011-08 A Type2  8  Type1  Type1 
12 2011-08 A Type2  8  Type1  Type2 
13 2011-08 A Type3 63  Type1  Type1 
14 2011-08 A Type3 63  Type1  Type2 
15 2011-08 A Type4  2  Type1  Type1 
16 2011-08 A Type4  2  Type1  Type2 
17 2012-03 A Type1 43  <NA>  Type1 
18 2012-03 A Type1 43  <NA>  Type2 
19 2011-01 B Type1  5  Type1  Type1 
20 2011-01 B Type2 89  Type1  Type1 
21 2011-01 B Type3  6  Type1  Type1 
22 2011-01 B Type4 25  Type1  Type1 
23 2011-08 B Type1 78 Type1 Type4  Type1 
24 2011-08 B Type2  2 Type1 Type4  Type1 
25 2011-08 B Type3 62 Type1 Type4  Type1 
26 2011-08 B Type4 46 Type1 Type4  Type1 
27 2012-03 B Type1  2  <NA>  Type1 
28 2011-01 C Type1  3  Type2  Type2 
29 2011-01 C Type2 24  Type2  Type2 
30 2011-01 C Type3 21  Type2  Type2 
31 2011-01 C Type4  8  Type2  Type2 
32 2011-08 C Type1  4  Type2  Type2 
33 2011-08 C Type2 13  Type2  Type2 
34 2011-08 C Type3 19  Type2  Type2 
35 2011-08 C Type4  3  Type2  Type2 
36 2012-03 C Type1  7  <NA>  Type2