subsetting based on number of observations in a factor variable

问题

how do you subset based on the number of observations of the levels of a factor variable? I have a dataset with 1,000,000 rows and nearly 3000 levels, and I want to subset out the levels with less say 200 observations.

data <- read.csv("~/Dropbox/Shared/data.csv", sep=";")

summary(as.factor(data$factor)
10001 10002 10003 10004 10005 10006 10007 10009 10010 10011 10012 10013 10014 10016        10017 10018 10019 10020 
  414   741  2202   205   159   591   194   678   581   774   778   738  1133   997   381   157   522     6 
10021 10022 10023 10024 10025 10026 10027 10028 10029 10030 10031 10032 10033 10034 10035 10036 10037 10038 
  398   416  1236   797   943   386   446   542   508   309   452   482   425   272   261   291   145   598 
10039 10040 10041 10043 10044 10065 10069 10075 10080 10104 10105 10106 10110 10112 10115 10117 10119 10121 
  119   263     9     9   179   390    70   465    19     3     7     5     4     1     1     1     2     6 
10123 10128 10150 10152 10154 10155 10168 10170 10173 10174 10176 10199 10210 10220 10240 10265 10270 10271 
    2   611     8     1     1     2    10     1     6     5     5     2     5     2     1     3     5     2

as you see from the summary, above, there are factors with only a few obs, and I want to remove the factors that have less than 100.

I tried the following, but it didn't work:

for (n in unique((data$factor))) {
    m<-subset(data, factor==n)
    o<-length(m[,1])
    data<-ifelse( o<100, subset(data, factor!=n), data)
}

回答1:

table, subset that, and match based on the names of that subset. Probably will want to droplevels thereafter.

EIDT

Some sample data:

set.seed(1234)
data <- data.frame(factor = factor(sample(10000:12999, 1000000, 
  TRUE, prob=rexp(3000))))

Has some categories with few cases

> min(table(data$factor))
[1] 1

Remove records from case with less than 100 of those with the same value of factor.

tbl <- table(data$factor)
data <- droplevels(data[data$factor %in% names(tbl)[tbl >= 100],,drop=FALSE])

Check:

> min(table(data$factor))
[1] 100

Note that data and factor are not very good names since they are also builtin functions.

回答2:

I figured it out using the following, as there is no reason to do things twice:

function (df, column, threshold) { 
    size <- nrow(df) 
    if (threshold < 1) threshold <- threshold * size 
    tab <- table(df[[column]]) 
    keep <- names(tab)[tab >  threshold] 
    drop <- names(tab)[tab <= threshold] 
    cat("Keep(",column,")",length(keep),"\n"); print(tab[keep]) 
    cat("Drop(",column,")",length(drop),"\n"); print(tab[drop]) 
    str(df) 
    df <- df[df[[column]] %in% keep, ] 
    str(df) 
    size1 <- nrow(df) 
    cat("Rows:",size,"-->",size1,"(dropped",100*(size-size1)/size,"%)\n") 
    df[[column]] <- factor(df[[column]], levels=keep) 
    df 
}

来源：https://stackoverflow.com/questions/10555858/subsetting-based-on-number-of-observations-in-a-factor-variable

标签

subset

r-factor