About

• N-gram https://en.wikipedia.org/wiki/N-gram
• Markov Chain
• Created for NaNoGenMo 2016 (National November Generated fiction Month)
• See NaNoGenMo2017 for this year's version

Setup

• Install Java 8+ JDK
• Install maven
• Install your favorite version of LaTeX (Either MikTeX or LiveTeX should work on Windows)

Basic Execution

Build app

• mvn test

Running generator

• steps...

Generating pdf document:

• mvn latex:latex
• Generate LaTeX output with mvn latex:latex
• requires "story.txt" in top directory (change path in \\input{} to change location)
• pdf file is in target/latex/book/book.pdf

(find mvn cmdline that runs all at once)

Suggested Commandline Arguments

• Word count
• Source filename
• Destination file (and/or directory?)
• IMarkov implementation (default to MarkovBigram)
• Formatted
• LaTeX does its own line formatting, so maybe just always format?

Next Tasks

• Reader reads from Internet
• implement 3-gram to get better sentences
• More unit tests
• Create main() that implements calls in testGenerate()
• half-done, methods moved out of MarkovBigramTest
• implement parsing of cmdline args

Bugs

• sometimes period appears after other punctuation
  • Create unit test demonstrating it

Ideas:

• html: hover display words that could have been chosen and probabilities for each
• variable-length sentences (would require bayesian-type methods?)

Questions

• should freqs be an instance field or input/returned making the class "pure functional" with no stored state?

---

Done

• better determination of proper nouns
• move docs to separate directory
• rename "First" to MarkovBigram
• have store() return freqs object instead of requiring it to be passed in
• Made freqs an instance field so it's not passed around
• find lib to do word-wrap
• Good old Apache Commons
• LaTeX output
• generates PDF file programmatically easier than using some pdf library, but the downside is it requires installation of a LaTeX distribution