setwd('/home/sek1ro/git/public/lab/ds/25-1/r')
df = read.csv("nbtrain.csv", stringsAsFactors = TRUE)
trdf = df[1:9010,]
tedf = df[9011:10010,]
remove(df)
library(e1071)

nb = naiveBayes(income ~ age + sex + educ, data = trdf, laplace = 1)
# p(A|B)*p(B) = p(AB) = p(B|A)*p(A)
# p(A|B) = p(B|A) * p(A) / p(B)
# апостер = услов * априор / маргин
nb$apriori / sum (nb$apriori)
nb$tables


pd = predict(nb, tedf)
(conf_mat = table(Actual = tedf$income, Predicted = pd))

conf_tot = function(conf_mat) {
  cat(1 - (sum(diag(conf_mat)) / sum(conf_mat)))
}

conf_class = function(conf_mat) {
  for (income in rownames(conf_mat)) {
    err = 1 - (conf_mat[income, income] / sum(conf_mat[income, ]))
    cat(sprintf("%s error %.2f%%\n", income, err * 100))
  }
}

conf_tot(conf_mat)
conf_class(conf_mat)

nb = naiveBayes(sex ~ age + educ + income, data = trdf, laplace = 1)
nb$apriori / sum (nb$apriori)
nb$tables

pd = predict(nb, tedf)
(conf_mat = table(Actual = tedf$sex, Predicted = pd))
conf_tot(conf_mat)
conf_class(conf_mat)



male = trdf[trdf$sex == "M", ]
female = trdf[trdf$sex == "F", ]

nbrandom = function() {
  mdf = male[sample(1:nrow(male), 3500),]
  fdf = female[sample(1:nrow(female), 3500), ]
  
  mfdf = rbind(mdf, fdf)
  
  mfnb = naiveBayes(sex ~ age + educ + income, data = mfdf, laplace = 1)
  mfnb$apriori / sum (mfnb$apriori)
  mfnb$tables
  
  mfpd = predict(mfnb, tedf)
  (mfconf_mat = table(Actual = tedf$sex, Predicted = mfpd))
  conf_tot(mfconf_mat)
  conf_class(mfconf_mat)
}

set.seed(Sys.time())
nbrandom()