// NYUSH F24 - gohai

​

let n = 4;

​

let lines; // array of individual lines

let input; // the entire text

let re;

let probabilities;

let generatedText = "";

​

let textMargin = 20;

let textX = textMargin;

let textY = textMargin;

​

function preload() {

  lines = loadStrings("text.txt");

  frameRate(5);

}

​

function setup() {

  createCanvas(400, 400);

  background(220);

  textSize(18);

​

  // this prepares a "regular expression"

  // used to split text into word-tokens

  let wordDelimiters = ' .,:;"“?!\n';

  re = new RegExp(

    "(?=[" + wordDelimiters + "])|(?<=[" + wordDelimiters + "])",

    "g"

  );

​

  input = lines.join(" ");

  probabilities = analyze(input);

  //console.log(probabilities);

}

​

function draw() {

  let nextChars = generate(probabilities, generatedText);

  //console.log(nextChars);

  generatedText = generatedText + nextChars;

​

  // print new text to canvas

  if (textX + textWidth(nextChars) >= width - textMargin) {

    textY = textY + textAscent() + textDescent();

    textX = textMargin;

  }

  text(nextChars, textX, textY);

  textX = textX + textWidth(nextChars);

}

​

function analyze(str) {

  // turn string into an array of word-tokens

  let words = str.split(re);

​

  let counts = {};

  // count how often each n-gram occurs

  for (let i = 0; i < words.length - n + 1; i++) {

    let ngram = "";

    for (let j = i; j < i + n; j++) {

      ngram = ngram + words[j];

    }

    if (counts[ngram]) {

      counts[ngram]++; // seen before, add one

    } else {

      counts[ngram] = 1; // start with count 1

    }

  }

  // turn the count into probability (0-1)

  for (let ngram in counts) {

    counts[ngram] = counts[ngram] / (words.length - n + 1);

  }

  return counts;

}

​

function generate(probabilities, previousOutput) {

  let selected;

  let needle = "";

​

  if (previousOutput.length == 0) {

    // at the beginning of generation, start with any

    selected = probabilities;

  } else {

    // for all other times, we need to first select

    // all the n-grams that start with the last tokens

    // of the current text

    let previousOutputWords = previousOutput.split(re);

​

    selected = {};

    needle = previousOutputWords.slice(-1 * (n - 1)).join("");

    //console.log('looking for', '"' + needle + '"')

    let sum = 0;

    for (let ngram in probabilities) {

      if (ngram.startsWith(needle)) {

        selected[ngram] = probabilities[ngram];

        sum = sum + probabilities[ngram];

      }

    }

    // normalize the probabilities

    for (let ngram in selected) {

      selected[ngram] = selected[ngram] / sum;

    }

    if (Object.keys(selected).length == 0) {

      console.error("This shouldn't happen");

    }

  }

​

  // pick a random number 0..1

  let rand = random();

  let sum = 0;

  for (let ngram in selected) {

    // check if we landed "inside" the probability

    // of the current character (higher probability

    // == more chance for a match)

    if (sum + selected[ngram] >= rand) {

      if (previousOutput.length == 0) {

        //console.log('picked', '"' + ngram + '"');

        return ngram; // return the entire ngram

      } else {

        //console.log('picked', '"' + ngram + '"');

        return ngram.substr(needle.length);

      }

    }

    sum = sum + selected[ngram];

  }

  console.error("This also shouldn't happen");

}

​