CODEWITHSUNDEEP
rfunctiondataframecomparison
Leprechault
Leprechault

Reputation: 1823

Data frame classification according to a rule

I have two data frames DF1 and DF2, DF2 is the comparison between some multiple treatments and p corresponding value. I would like to create a variable groups in DF2 depending on the significance of p, were my final output is:

# 
   trat      sp1      sp2      sp3      sp4    groups 
1 P12A 4.653732 2.490977 4.236323 5.382113 "P12AvsP12BvsP12CvsP12D" 
2 P12A 5.009581 2.254713 4.604529 4.842553 "P12AvsP12BvsP12CvsP12D" 
3 P12A 4.809242 2.391435 4.675318 4.732977 "P12AvsP12BvsP12CvsP12D" 
4 P12A 5.053077 2.483129 4.561690 5.311215 "P12AvsP12BvsP12CvsP12D" 
5 P12A 5.356384 2.745474 4.616074 5.114969 "P12AvsP12BvsP12CvsP12D" 
6 P12A 5.186120 2.384487 4.401133 5.041926 "P12AvsP12BvsP12CvsP12D" 
7  P12A 5.175057 2.443725 4.678268 4.931249 "P12AvsP12BvsP12CvsP12D" 
8  P12A 4.942661 2.288388 4.796253 5.123081 "P12AvsP12BvsP12CvsP12D" 
9  P12A 5.049273 2.518262 4.022213 5.334939 "P12AvsP12BvsP12CvsP12D" 
10 P12A 5.301788 2.431241 4.778733 4.880814 "P12AvsP12BvsP12CvsP12D" 
11 P12A 5.079584 2.403393 4.882675 4.625801 "P12AvsP12BvsP12CvsP12D" 
12 P12A 4.644050 2.488449 4.712999 4.699644 "P12AvsP12BvsP12CvsP12D" 
13 P12A 4.898727 2.293527 5.275999 4.898135 "P12AvsP12BvsP12CvsP12D" 
14 P12A 5.248847 2.286491 4.927688 4.763063 "P12AvsP12BvsP12CvsP12D" 
15 P12A 5.239704 2.529684 4.930668 5.212917 "P12AvsP12BvsP12CvsP12D" 
16 P13A 5.229522 2.570899 4.781054 4.852892 "P13A"
17 P13A 4.962258 2.982522 5.000820 4.659382 "P13A"
18 P13A 4.694233 2.214868 5.066398 4.887591 "P13A"
19 P13A 4.782794 2.169470 4.968994 4.839169 "P13A"
20 P13A 4.739379 2.807208 4.967769 4.872314  "P13A"
21 P13A 4.724091 2.362886 5.074778 5.064301  "P13A"
22 P13A 5.111046 2.468498 5.004575 4.895111  "P13A"
23 P13A 4.948762 2.671930 4.712485 4.761943  "P13A"
24 P13A 5.418908 2.676300 4.892651 5.128098  "P13A"
25 P13A 4.967169 2.700370 5.212501 4.563933  "P13A"
26 P13A 4.950029 2.384279 4.946293 4.875622  "P13A"
27 P13A 5.013728 2.278861 4.845646 5.191396  "P13A"
#

That is, if p = or> 0.05, I'm joining the levels of treatment ("P12AvsP12BvsP12CvsP12D") separated by vs, otherwise I repeat the name of the level ("P13A"), follows my CRM below and my function that doesn't work, thank you,

## Data frame 1 
pares<-c("P12A vs P12B","P12A vs P12C","P12A vs P12D","P12A vs 
P12E","P16A vs P12F", 
"P18A vs P12G","P20A vs P12H","P21A vs P12I","P22A vs P13A","P30A vs 
P13B","P33A vs P142", 
"P34A vs p142","P35A vs P148","P35A vs p148") 
p<-c(1,1,4.00E-04,1.00E-04,0.0272,1,0.0012,1,2.00E-04,0.0281,2.00E-04,1,4.00E-04,1) 
DF1<-data.frame(pares,p) 
head(DF1) 

#Data frame 2 
#Factor 
trat <- gl(3, 15, labels = paste("P", 12:14, "A",sep="")) 
# Response variable 
set.seed(124) 
sp  <- cbind(c(rnorm(10,  5, 0.25), rnorm(35, 5, 0.25)), rnorm(45, 2.5, 
0.25), 
              c(rnorm(10, 4.5, 0.25), rnorm(35, 5, 0.25)), rnorm(45, 5, 
0.25)) 
colnames(sp) <- c("sp1", "sp2", "sp3", "sp4") 
DF2<-data.frame(trat,sp) 

## Classification

# Function tentative
         if (DF1$p>=0.05) { 
         DF2$groups = DF1$pares 
} else { 
         DF2$groups = DF1$trat 
} 

head(DF2) 
#

Upvotes: 1

Views: 66

Answers (1)

tfc
tfc

Reputation: 596

One approach is to iterate over each treatment filtering the comparisons among treatments that are above the threshold. Having the mapping of treatment => valid_pairs it is easy to create the groups column. Below is the commented code:

combine_pairs <- function(trat, threshold = 0.05){
  #find every pair related to trat (it can be "trat vs X" or "X vs trat")
  #this can be expressed using regular expressions as "(^trat vs|vs trat$)"
  pares <- grepl(paste0("(^", trat, " vs|vs ", trat, "$)"), DF1$pares)
  #among those pairs find the names of the ones that are above the threshold
  matches <- DF1$pares[pares & DF1$p >= threshold]
  if (length(matches) >= 1){
    #clean up string for output, removing the "vs" and the trat name in each pair
    trat_names <- gsub(paste0("( vs |", trat, ")"), "", matches)
    groups <- paste0(trat, "vs", paste(trat_names, collapse = "vs"))
  }else{
    groups <- trat #what to show when no comparison is >= threshold
  }

  groups
}

#create a named list with the correct groups value for each treatment 
groups <- setNames(lapply(levels(DF2$trat), combine_pairs), levels(DF2$trat))
DF2$groups <- groups[DF2$trat]

The output is a little bit different than in your example (P12D is not listed for treatment P12A). In your example input the p value for this pair (P12A vs P12D) is 0.0004.

Upvotes: 1

Related Questions