using System;
using System.Collections.Generic;
using System.Linq;
using MarkovGrams.Utilities;
namespace MarkovGrams
{
///
/// An unweighted character-based markov chain.
///
public class UnweightedMarkovChain
{
///
/// The random number generator
///
private Random rand = new Random();
///
/// The ngrams that this markov chain currently contains.
///
private List ngrams;
///
/// Whether to always start generating a new word from an n-gram that starts with
/// an uppercase letter.
///
public bool StartOnUppercase = false;
///
/// The generation mode to use when running the Markov Chain.
///
///
/// The input n-grams must have been generated using the same mode specified here.
///
public GenerationMode Mode { get; private set; } = GenerationMode.CharacterLevel;
///
/// Creates a new character-based markov chain.
///
/// The ngrams to populate the new markov chain with.
public UnweightedMarkovChain(IEnumerable inNgrams, GenerationMode inMode)
{
ngrams = new List(inNgrams);
Mode = inMode;
}
///
/// Returns a random ngram that's currently loaded into this UnweightedMarkovChain.
///
/// A random ngram from this UnweightMarkovChain's cache of ngrams.
public string RandomNgram()
{
IEnumerable validNGrams = StartOnUppercase ? ngrams.Where((ngram) => char.IsUpper(ngram[0])) : ngrams;
if (validNGrams.Count() == 0)
throw new Exception($"Error: No valid starting ngrams were found (StartOnUppercase: {StartOnUppercase}).");
return validNGrams.ElementAt(rand.Next(0, validNGrams.Count()));
}
///
/// Generates a new random string from the currently stored ngrams.
///
///
/// The length of ngram to generate.
/// Note that this is a target, not a fixed value - e.g. passing 2 when the n-gram order is 3 will
/// result in a string of length 3. Also, depending on the current ngrams this markov chain contains,
/// it may end up being cut short.
///
/// A new random string.
public string Generate(int length)
{
return Generate(length, out float noop);
}
public string Generate(int length, out float choicePointRatio)
{
string result = RandomNgram();
string lastNgram = result;
List choiceCounts = new List(); int i = 0;
while((Mode == GenerationMode.CharacterLevel ? result.Length : result.Split(' ').Length) < length)
{
// The substring that the next ngram in the chain needs to start with
string nextStartsWith = Mode == GenerationMode.CharacterLevel ? lastNgram.Substring(1) : string.Join(" ", lastNgram.Split(' ').Skip(1));
// Get a list of possible n-grams we could choose from next
List nextNgrams = ngrams.FindAll(gram => gram.StartsWithFast(nextStartsWith));
choiceCounts.Add(nextNgrams.Count);
// If there aren't any choices left, we can't exactly keep adding to the new string any more :-(
if(nextNgrams.Count == 0)
break;
// Pick a random n-gram from the list
string nextNgram = nextNgrams.ElementAt(rand.Next(0, nextNgrams.Count));
// Add the last character from the n-gram to the string we're building
if (Mode == GenerationMode.CharacterLevel)
result += nextNgram[nextNgram.Length - 1];
else
result += ' ' + nextNgram.Split(' ').Last();
lastNgram = nextNgram; i++;
}
if (choiceCounts.Sum() > 0)
choicePointRatio = (float)choiceCounts.Sum() / (float)(i + 1);
else
choicePointRatio = 0;
return result;
}
}
}