ham.js

// take the last 100 tweets from the user
{ 
  accountAge: 4,
  mins: 2880, // number of mins to get to 100 tweets
  frequentLen: 4, // When they retweet the same account more than 4 times per hundred ++
  totalFreq: 81, // Total count of above
  convos: 3, // number of tweets directed at another user
  listed_count: 23, // how often they are listed
  numOriginalTweets: 9, // how many of the 100 are created by the tweeter
  sourcesLen: 3 // how many different platforms (iOS, android, web) they tweeted from
}


const bayes = require('node-bayes')
const { firebase } = require('./config')
const _ = require('underscore')

const TRAINING_COLUMNS = ['accountAge', 'mins', 'frequentLen', 'totalFreq', 'created',
              'convos', 'listed_count', 'numOriginalTweets', 'sourcesLen']
let cls

const setup = async () => {
  let ham, spam
  
  await firebase
    .ref('ham')
    .once('value')
    .then(snap => ham = snap.val())

  await firebase
    .ref('spam')
    .once('value')
    .then(snap => spam = snap.val())

  const spamArr = _.shuffle(Object.keys(spam))
                     .slice(0, 50)
                     .map(key => spam[key])

  const hamArr = Object.keys(ham).map(key => ham[key])

  const buildArray = (arr, tag) => {
      return arr.map(obj => {
        delete obj.created
        return [...Object.values(obj), tag]
    })
  }

  const TRAINING_DATA = [...buildArray(hamArr, 'ham'), ...buildArray(spamArr, 'spam')]
  

  cls = new bayes.NaiveBayes({
    columns: TRAINING_COLUMNS,
    data: TRAINING_DATA,
    verbose: true
  })

  cls.train()
}

setup()

exports.twitterAlgo = (ranker, screen_name) => {
  const answer = cls.predict(Object.values(ranker))
  console.log(answer, screen_name)
}