如果满足条件，如何保留一行并删除其他

Question

I am working with taxonomic data and have gotten my data to the second last step before I can display it graphically. However, I need rows to match conditions and this is where I am stuck - well stuck because I do not want to do it manually.
My data:

x <- data.frame("Phylum" = c("Chordata", "Chordata", "Chordata", "Chordata", "Chordata", "Chordata"),
                "Class" = c("NA", "Actinopterygii", "Actinopterygii", "Actinopterygii", "Actinopterygii", "Actinopterygii"),
                "Order" = c("NA", "NA", "Gadiformes", "Gadiformes", "Gadiformes", "Gadiformes"), 
                "Family" = c("NA", "NA", "NA", "Moridae", "Moridae", "Moridae"), 
                "Genus" = c("NA", "NA", "NA", "NA", "Notophycis", "Notophycis"), 
                "Species" = c("NA", "NA", "NA", "NA", "NA", "Notophycis marginata"),
                 Number = c(21616, 12123, 1497, 730,730,730))

The wanted end result:

y <- data.frame("Phylum" = c("Chordata", "Chordata", "Chordata", "Chordata"), 
                "Class" = c("NA", "Actinopterygii", "Actinopterygii", "Actinopterygii"), 
                "Order" = c("NA", "NA", "Gadiformes", "Gadiformes"), "Family" = c("NA", "NA", "NA", "Moridae"), 
                "Genus" = c("NA", "NA", "NA", "Notophycis"), "Species" = c("NA", "NA", "NA", "Notophycis marginata"), 
                 Number = c(9493, 10626, 767, 730))

This is a simple subset example from a much larger more complicated dataset. So if I could put this into code somehow:

• sum of Number ( Phylum == "P1" & Class == "NA" ) - sum of Number ( Class == "C1" & Order == "NA" ) IF phylum matches and this would equal P1's new Number
• sum of Number ( Class == "C1" & Order== "NA" ) - sum of Number ( Order == "O1" & Family == "NA" ) IF class matches and this would equal C1's new Number etc...

BUT if the Number matches for multiple rows I need to have code to evaluate those rows and choose the row that has the least amount of NAs and keep that Number...

I would presume I am looking to code a function to do this, but have no idea where to even start!

Appreciate the help :)

UPDATE

Tester:

Phylum  Class   Order   Family  Genus   Species Reads_sum
Chordata    Elasmobranchii  Carcharhiniformes   NA  NA  NA  31
Chordata    Actinopterygii  Perciformes Scombridae  NA  NA  589
Chordata    Elasmobranchii  Carcharhiniformes   Pentanchidae    NA  NA  31
Chordata    Actinopterygii  Myctophiformes  Myctophidae Notoscopelus    NA  208
Chordata    Actinopterygii  Perciformes Scombridae  Katsuwonus  NA  589
Chordata    Actinopterygii  Myctophiformes  Myctophidae Notoscopelus    Notoscopelus caudispinosus  178
Chordata    Actinopterygii  Perciformes Scombridae  Katsuwonus  Katsuwonus pelamis  589
Cnidaria    Hydrozoa    Leptothecata    Plumulariidae   NA  NA  69
Cnidaria    Hydrozoa    Leptothecata    Plumulariidae   Plumularia  NA  69
Echinodermata   Ophiuroidea NA  NA  NA  NA  146
Echinodermata   Ophiuroidea Ophiurida   NA  NA  NA  137
Echinodermata   Ophiuroidea Ophiurida   Ophiuridae  NA  NA  137
Echinodermata   Ophiuroidea Ophiurida   Ophiuridae  Ophioplinthus   NA  137
Echinodermata   Ophiuroidea Ophiurida   Ophiuridae  Ophioplinthus   Ophioplinthus accomodata    137
Mollusca    Cephalopoda Oegopsida   Ommastrephidae  NA  NA  34311
Ochrophyta  Phaeophyceae    Ectocarpales    Acinetosporaceae    NA  NA  29

Code that preforms what I would like BUT would have to change the variables each time:

Tester$Reads_sum[Tester$Class == "Ophiuroidea" & Tester$Order == "NA"] - sum(Tester$Reads_sum[Tester$Class == "Ophiuroidea" & Tester$Order != "NA" & Tester$Family == "NA"])

And so I was hoping something like this would work and I would just need to change Class to other selected taxonomic ranks:

for (i in unique(Tester$Class)){
  Tester$Test.1 <- ifelse(Tester$Class != "NA" & Tester$Order == "NA", 
                           Tester$Reads_sum[Tester$Class == i & Tester$Order == "NA"] - sum(Tester$Reads_sum[Tester$Class == i & Tester$Order != "NA" & Tester$Family == "NA"]), 0)
  }

But it is giving me an NA instead of 9.

The end data should look like this:

Phylum  Class   Order   Family  Genus   Species Reads_sum
Chordata    Elasmobranchii  Carcharhiniformes   Pentanchidae    NA  NA  31
Chordata    Actinopterygii  Myctophiformes  Myctophidae Notoscopelus    NA  30
Chordata    Actinopterygii  Myctophiformes  Myctophidae Notoscopelus    Notoscopelus caudispinosus  178
Chordata    Actinopterygii  Perciformes Scombridae  Katsuwonus  Katsuwonus pelamis  589
Cnidaria    Hydrozoa    Leptothecata    Plumulariidae   Plumularia  NA  69
Echinodermata   Ophiuroidea NA  NA  NA  NA  9
Echinodermata   Ophiuroidea Ophiurida   Ophiuridae  Ophioplinthus   Ophioplinthus accomodata    137
Mollusca    Cephalopoda Oegopsida   Ommastrephidae  NA  NA  34311
Ochrophyta  Phaeophyceae    Ectocarpales    Acinetosporaceae    NA  NA  29

Answer 1

thanks for the update. I've come up with something I think meets what you are looking for, but will need some support.

Am I right in thinking its a tree like data in the order c("Phylum", "Class", "Order", "Family", "Genus", "Species") ? and that you are interested in finding for each level of the tree, you want to remove the values of the layer below?

I hope my code isn't too confusing, I found the data challenging to use in its current format. I prefer to split it in to the levels of the tree, ie the ones that just have Phylum data, all the way down to the ones that have all levels of the tree. To do so, I am most comfortable using the data.table package.

I've used lapply's where I can as I find them easy to interpret once you use them a lot. I'm sure there will be a more efficient solution out there, but as a starter, I think knowing and understanding the steps required are more important.

# using data.table package, as I find it quicker and easier to work with 
# for complex problems. Run the hashed out command below if you dont have it
# install.packages("data.table")
library(data.table)

# turning in to a data.table, similar to data.frame, but some differences.
dt <- as.data.table(Tester)
# I am making an id, which I will use to split up this data. Different rows 
# have different structures, as its a tree structure, so I am going to break
# the data up
dt[, id := 1:.N]

# to do so i need to know the order of significance of the tree. I believe
# they go in this order:
col_structure <- c("Phylum", "Class", "Order", "Family", "Genus", "Species")

# I want to find out at which level of the tree each row is, so I am going
# to change teh shape from wide to long, and then do some row aggregation on 
# the single column, to group
melt_dt <- melt(dt, id.vars = "id", 
                measure.vars = col_structure)
# tip: try not to use "NA", but instead NA, they have different structures 
# and built in commands like is.na make them easier to differentiate
melt_dt[value == "NA", value := NA]
melt_dt <- melt_dt[!is.na(value)]
melt_dt[]
# using a data.table command .N, grouped by id, to find out how many non NA
# values there are, this will tell me where it is in the tree
group_ids <- melt_dt[, .N, by = id]

# Ok, so now I will split up each row in to where it sits in the tree
split_ids <- split(group_ids, group_ids$N)
split_ids
# pull out the number of levels of tree for easy use
levels <- seq_along(split_ids)

# merge back in the original data, so we have the same data at the start, but
# split up in to new sets. Makes it easier to think about the problem
split_dt <- lapply(levels, function(x){
  out <- merge(split_ids[[x]], dt, by = "id")
  N <- as.numeric(names(split_ids)[x])
  # using keys in my data, to make easy extraction. means rather than do
  # Phylum == "a" & Class == "b" later on, if Phylum & Class are the keys,
  # then can use command J("a", "b"). See next stage
  setkeyv(out, col_structure[1:N])
  out
})

# Now I'm going to add the value in. I will look at the next level of the tree
# and remove the values from that level from the reads_sum. Try it with setting
# x = 1.
# I've removed bottom element of the tree, don't know what to do with them
split_dt_with_value <- lapply(levels[1:(length(levels)-1)], function(x){
  # similar to for loop, but using data.table keys to extract data
  out <- split_dt[[x]]
  out$Test.1 <- out$Reads_sum - sapply(1:nrow(out), function(i){
    sum(split_dt[[(x+1)]][J(out[i, key(out), with = FALSE])]$Reads_sum,
        na.rm = TRUE)
  })
  out
})

# combine results, and with the bottom tree level
combined <- rbindlist(c(split_dt_with_value,
                        split_dt[max(levels)]), 
                        fill = TRUE)
# turn it back in to data frame form 
combined <- as.data.frame(combined)
combined

please have a look and let me know if any steps are confusing, or any of the logic is incorrect :)

Cheers, Jonny