simple bayesian spam classifier using laplace smoothing. quickly thrown together but i'll try to polish this tomorrow.

spam.rb

def sum(arr)
  res = 0
  arr.each{|e| res+=e}
  res
end

def mult(arr)
  res = 1
  arr.each{|e| res*=e}
  res
end

$training = {
  :spam=>["offer is secret", "click secret link", "secret sports link"],
  :ham=>["play sports today", "went play sports", "secret sports event", "sports is today", "sports costs money"]
}

$LSk = 1.0
$dict = {}
$global_dict = {}
$global_dict.default=0.0
$priors={}
$total_words={}
$total_words.default=0
$total_words_all=0
total_messages=0
$training.each_value{|messages| total_messages += messages.size}

$training.each_pair{|klass, messages|
  $dict[klass] = {}
  $dict[klass].default=0.0
  messages.each{|msg|
    msg.split.each{|word|
      $dict[klass][word]+=1
      $global_dict[word]+=1
      $total_words[klass]+=1
    }
  }
  $total_words_all+=$total_words[klass]
  $priors[klass] = (messages.size.to_f + $LSk) / (total_messages.to_f + $LSk * $training.keys.size)
}

p $dict
p $priors

def prob(word, klass)
  ($dict[klass][word] + $LSk) / ($total_words[klass] + $LSk * $global_dict.size)
end

def classify(message)
  res={}
  $training.each_key{|klass|
    words = message.split
    num = mult(words.map{|w| prob(w, klass)}) * $priors[klass]
    denom = 0.0 
    $training.each_key{|kk|
      denom += mult(words.map{|w| prob(w, kk)}) * $priors[kk]
    }
    res[klass] = num/denom
  }
  res
end

puts prob("today", :spam)
puts prob("today", :ham)

p classify("today is secret")