sample program for category data analysis, made for a challenge in Kaggle.

prediction1.R

library(dplyr)
library(ggplot2)
library(reshape2)

train <- read.csv("train.csv")
#test <- read.csv("test_v2.csv")

az <- train %.%
  filter(record_type==1) %.%
  select(A:G)

cramer.coe <- function(dat){
  row.sum <- apply(dat, 1, sum)
  col.sum <- apply(dat, 2, sum)
  row.length <- length(dat[,1])
  col.length <- length(dat[1,])
  my.sum = 0
  for(i in 1:row.length){
    for(j in  1:col.length){
      num <- dat[i,j]^2 / (row.sum[i] * col.sum[j])
      my.sum <- my.sum + num
    }
  }
  coe <- sqrt((my.sum - 1) / (row.length - 1))
  coe
}

dcast.coe <- function(dat,form) {
  d <- dcast(dat,form,length,value.var=1)
  d.m <- matrix(as.double(as.matrix(d[,2:ncol(d)])),nrow(d),ncol(d)-1)
  cramer.coe(d.m)
}

cols1 <- names(az)
cols2 <- names(az)
df <- data.frame()
for (c1 in cols1) for (c2 in cols2) {
  t <- paste(c1,"~",c2)
  coe <- if (c1==c2) NA else dcast.coe(az,formula(t))
  #cat(paste(t,coe,"\n"))
  df <- rbind(df,data.frame(c1=c1,c2=c2,coe=coe))
}
p <- ggplot(df,aes(c1,c2))+geom_tile(aes(fill=coe),color="white")+
  scale_fill_gradient(low="white",high="steelblue")
print(p)