smartround: Smart round of variables

Description

Smart round of variables

Usage

smartround(x, ndistinct = 100, freq = ~1, stats = c("mean", "median"),
  method = c("centroid", "median", "ward"), short = FALSE)

Arguments

a numeric matrix or data frame. It must contains the variables in the freq formula

ndistinct

number of distinct values you want to obtain. If there are missing values, you will get ndistinct + 1 distint values.

freq

a one-sided, single term formula specifying frequency weights

stats

statistic to compute each distinct value

method

method to calculate the different groups. See hclustvfreq

short

logical. If it is TRUE, the function returns the data collected in a frequency table. Otherwise, it returns a object with the same rows than the original data.

Value

If short is TRUE, it returns a frequency table of dimension ndistinct, and missing values are removed.
Otherwise, a vector or a data frame with the same rows than the original data is returned. It preserves the original order.

Details

If you want to reduce the number of unique elements in your data set to a specific number, you can use this function. It collects the numeric data (uni or multi-dimensional) into groups and estimates the values of the centers for each group.

Examples

Run this code

smartround(c(1:10,NA),2)
smartround(c(1:10,NA),2,short=TRUE)
smartround(iris[,1:4],7,short=TRUE)

if(require(hflights)){

  x <- hflights[,c("ArrDelay")]
  xr <- smartround(x, 100)
  print(length(unique(xr))) ## 101: 100 and NA
  ## xr is almost equal to x
  print(cor(x,xr,use="pairwise.complete.obs")) # 0.998

  ## Now, with a 2-dimension data frame

  d0 <- hflights[,c("ArrDelay","DepDelay")]
  t0 <- tablefreq(d0)
  print(nrow(t0)) # there are 11046 distinct cases

  ## we reduce to just 100  distinct cases
  d2 <- smartround(d0,100)
  ## The correlation is greater than 0.96
  print(cor(cbind(d0,d2),use="pairwise.complete.obs"))

  print(system.time(t1 <- smartround(d0,100,short=TRUE)))
  ## this is fast
  print(system.time(tfast <- smartround(t0,100,freq=~freq, short=TRUE)))
  print(all.equal(t1,tfast))
  print(nrow(t1))
}