Implement word-level ngrams

MarkovGrams/NGrams.cs

@@ -4,6 +4,12 @@ using System.Linq;
 
 namespace MarkovGrams
 {
+	public enum GenerationMode
+	{
+		CharacterLevel,
+		WordLevel
+	}
+
 	/// <summary>
 	/// A collection of methods to generate various different types of n-grams.
 	/// </summary>
@@ -15,12 +21,12 @@ namespace MarkovGrams
 		/// <param name="words">The words to turn into n-grams.</param>
 		/// <param name="order">The order of n-gram to generate..</param>
 		/// <returns>A unique list of n-grams found in the given list of words.</returns>
-		public static IEnumerable<string> GenerateFlat(IEnumerable<string> words, int order, bool distinct = true)
+		public static IEnumerable<string> GenerateFlat(IEnumerable<string> words, int order, GenerationMode mode, bool distinct = true)
 		{
 			List<string> results = new List<string>();
 			foreach (string word in words)
 			{
-				results.AddRange(GenerateFlat(word, order));
+				results.AddRange(GenerateFlat(word, order, mode));
 			}
 			if (distinct) return results.Distinct();
 			return results;
@@ -32,13 +38,18 @@ namespace MarkovGrams
 		/// <param name="str">The string to n-gram-ise.</param>
 		/// <param name="order">The order of n-gram to generate.</param>
 		/// <returns>A unique list of n-grams found in the specified string.</returns>
-		public static IEnumerable<string> GenerateFlat(string str, int order)
+		public static IEnumerable<string> GenerateFlat(string str, int order, GenerationMode mode)
 		{
 			List<string> results = new List<string>();
-			for(int i = 0; i < str.Length - order; i++)
+			if (mode == GenerationMode.CharacterLevel) {
-			{
+				for (int i = 0; i < str.Length - order; i++)
 					results.Add(str.Substring(i, order));
 			}
+			else {
+				string[] parts = str.Split(" ".ToCharArray());
+				for (int i = 0; i < parts.Length; i++)
+					results.Add(string.Join(" ", str.Skip(i).Take(order)));
+			}
 			return results.Distinct();
 		}
 
@@ -49,19 +60,11 @@ namespace MarkovGrams
 		/// <param name="words">The words to n-gram-ise.</param>
 		/// <param name="order">The order of ngrams to generate.</param>
 		/// <returns>The weighted dictionary of ngrams.</returns>
-		public static Dictionary<string, int> GenerateWeighted(IEnumerable<string> words, int order)
+		public static Dictionary<string, int> GenerateWeighted(IEnumerable<string> words, int order, GenerationMode mode)
 		{
 			Dictionary<string, int> results = new Dictionary<string, int>();
 			foreach(string word in words)
-			{
+				GenerateWeighted(word, order, mode);
-				Dictionary<string, int> wordNgrams = GenerateWeighted(word, order);
-				foreach(KeyValuePair<string, int> ngram in wordNgrams)
-				{
-					if(!results.ContainsKey(ngram.Key))
-						results[ngram.Key] = 0;
-					results[ngram.Key] += ngram.Value;
-				}
-			}
 			return results;
 		}
 		/// <summary>
@@ -70,17 +73,16 @@ namespace MarkovGrams
 		/// <param name="str">The string to n-gram-ise.</param>
 		/// <param name="order">The order of n-grams to generate.</param>
 		/// <returns>The weighted dictionary of ngrams.</returns>
-		public static Dictionary<string, int> GenerateWeighted(string str, int order)
+		public static void GenerateWeighted(string str, int order, GenerationMode mode, Dictionary<string, int> results)
 		{
-			Dictionary<string, int> results = new Dictionary<string, int>();
+			string[] parts = mode == GenerationMode.WordLevel ? str.Split(" ".ToCharArray()) : null;
 			for(int i = 0; i < str.Length - order; i++)
 			{
-				string ngram = str.Substring(i, order);
+				string ngram = mode == GenerationMode.CharacterLevel ? str.Substring(i, order) : string.Join(" ", parts.Skip(i).Take(order));
 				if(!results.ContainsKey(ngram))
 					results[ngram] = 0;
 				results[ngram]++;
 			}
-			return results;
 		}
 	}
 }

MarkovGrams/Program.cs

@@ -20,7 +20,8 @@ namespace MarkovGrams
 	{
 		public static int Main(string[] args)
 		{
-			Mode mode = Mode.None;
+			Mode operationMode = Mode.None;
+			GenerationMode generationMode = GenerationMode.CharacterLevel;
 			List<string> extras = new List<string>();
 			StreamReader wordlistSource = new StreamReader(Console.OpenStandardInput());
 			int order = 3, length = 8, count = 10;
@@ -62,8 +63,11 @@ namespace MarkovGrams
 					case "start-uppercase":
 						startOnUppercase = true;
 						break;
+					case "words":
+						generationMode = GenerationMode.WordLevel;
+						break;
 					case "help":
-						mode = Mode.Help;
+						operationMode = Mode.Help;
 						break;
 					default:
 						Console.Error.WriteLine($"Error: Unknown option '{args[i]}'.");
@@ -71,8 +75,8 @@ namespace MarkovGrams
 				}
 			}
 
-			if(mode != Mode.Help && extras.Count > 0)
+			if(operationMode != Mode.Help && extras.Count > 0)
-				mode = (Mode)Enum.Parse(typeof(Mode), extras.ShiftAt(0).Replace("markov-w", "weightedmarkov"), true);
+				operationMode = (Mode)Enum.Parse(typeof(Mode), extras.ShiftAt(0).Replace("markov-w", "weightedmarkov"), true);
 
 
 			// ------------------------------------------------------------------------------------------
@@ -86,12 +90,13 @@ namespace MarkovGrams
 				return new string[] { word.Trim() };
 			});
 
-			switch (mode)
+			switch (operationMode)
 			{
 				case Mode.Markov:
 					Stopwatch utimer = Stopwatch.StartNew();
 					UnweightedMarkovChain unweightedChain = new UnweightedMarkovChain(
-						NGrams.GenerateFlat(words, order)
+						NGrams.GenerateFlat(words, order, generationMode),
+						generationMode
 					);
 					unweightedChain.StartOnUppercase = startOnUppercase;
 
@@ -103,7 +108,8 @@ namespace MarkovGrams
 				case Mode.WeightedMarkov:
 					Stopwatch wtimer = Stopwatch.StartNew();
 					WeightedMarkovChain weightedChain = new WeightedMarkovChain(
-						NGrams.GenerateWeighted(words, order)
+						NGrams.GenerateWeighted(words, order, generationMode),
+						generationMode
 					);
 					weightedChain.StartOnUppercase = startOnUppercase;
 
@@ -113,7 +119,7 @@ namespace MarkovGrams
 					break;
 
 				case Mode.NGrams:
-					foreach (string ngram in NGrams.GenerateFlat(words, order, ngramsUnique))
+					foreach (string ngram in NGrams.GenerateFlat(words, order, generationMode, ngramsUnique))
 						Console.WriteLine(ngram);
 
 					break;
@@ -137,6 +143,7 @@ namespace MarkovGrams
 					Console.WriteLine("    --order {number}        Use the specified order when generating n-grams (default: 3)");
 					Console.WriteLine("    --length {number}       The target length of word to generate (Not available in ngrams mode)");
 					Console.WriteLine("    --count {number}        The number of words to generate (Not valid in ngrams mode)");
+					Console.WriteLine("    --words                 Generate ngrams on word-level instead of character-level (Applies to all modes)");
 					Console.WriteLine("    --no-split              Don't split input words on whitespace - treat each line as a single word");
 					Console.WriteLine("    --lowercase             Convert the input to lowercase before processing");
 					Console.WriteLine("    --start-uppercase       Start the generating a word only with n-grams that start with a capital letter");

MarkovGrams/UnweightedMarkovChain.cs

@@ -25,13 +25,22 @@ namespace MarkovGrams
 		/// </summary>
 		public bool StartOnUppercase = false;
 
+		/// <summary>
+		/// The generation mode to use when running the Markov Chain.
+		/// </summary>
+		/// <remarks>
+		/// The input n-grams must have been generated using the same mode specified here.
+		/// </remarks>
+		public GenerationMode Mode { get; private set; } = GenerationMode.CharacterLevel;
+
 		/// <summary>
 		/// Creates a new character-based markov chain.
 		/// </summary>
 		/// <param name="inNgrams">The ngrams to populate the new markov chain with.</param>
-		public UnweightedMarkovChain(IEnumerable<string> inNgrams)
+		public UnweightedMarkovChain(IEnumerable<string> inNgrams, GenerationMode inMode)
 		{
 			ngrams = new List<string>(inNgrams);
+			Mode = inMode;
 		}
 
 		/// <summary>
@@ -63,7 +72,7 @@ namespace MarkovGrams
 			while(result.Length < length)
 			{
 				// The substring that the next ngram in the chain needs to start with
-				string nextStartsWith = lastNgram.Substring(1);
+				string nextStartsWith = Mode == GenerationMode.CharacterLevel ? lastNgram.Substring(1) : lastNgram.Split(' ')[0];
 				// Get a list of possible n-grams we could choose from next
 				List<string> nextNgrams = ngrams.FindAll(gram => gram.StartsWith(nextStartsWith));
 				// If there aren't any choices left, we can't exactly keep adding to the new string any more :-(
@@ -72,7 +81,10 @@ namespace MarkovGrams
 				// 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 += string.Join(" ", nextNgram.Split(' ').Skip(1));
 				lastNgram = nextNgram;
 			}
 

MarkovGrams/WeightedMarkovChain.cs

@@ -24,17 +24,27 @@ namespace MarkovGrams
 		/// </summary>
 		public bool StartOnUppercase = false;
 
+		/// <summary>
+		/// The generation mode to use when running the Markov Chain.
+		/// </summary>
+		/// <remarks>
+		/// The input n-grams must have been generated using the same mode specified here.
+		/// </remarks>
+		public GenerationMode Mode { get; private set; } = GenerationMode.CharacterLevel;
+
 		/// <summary>
 		/// Creates a new character-based markov chain.
 		/// </summary>
 		/// <param name="inNgrams">The ngrams to populate the new markov chain with.</param>
-		public WeightedMarkovChain(Dictionary<string, double> inNgrams) {
+		public WeightedMarkovChain(Dictionary<string, double> inNgrams, GenerationMode inMode) {
 			ngrams = inNgrams;
+			Mode = inMode;
 		}
-		public WeightedMarkovChain(Dictionary<string, int> inNgrams) {
+		public WeightedMarkovChain(Dictionary<string, int> inNgrams, GenerationMode inMode) {
 			ngrams = new Dictionary<string, double>();
 			foreach (KeyValuePair<string, int> ngram in inNgrams)
 				ngrams[ngram.Key] = ngram.Value;
+			Mode = inMode;
 		}
 
 		/// <summary>
@@ -77,7 +87,7 @@ namespace MarkovGrams
 			{
 				wrandom.ClearContents();
 				// The substring that the next ngram in the chain needs to start with
-				string nextStartsWith = lastNgram.Substring(1);
+				string nextStartsWith = Mode == GenerationMode.CharacterLevel ? lastNgram.Substring(1) : lastNgram.Split(' ')[0];
 				// Get a list of possible n-grams we could choose from next
 				Dictionary<string, double> convNextNgrams = new Dictionary<string, double>();
 				ngrams.Where(gram_data => gram_data.Key.StartsWith(nextStartsWith))
@@ -89,7 +99,10 @@ namespace MarkovGrams
 				// Pick a random n-gram from the list
 				string nextNgram = wrandom.Next();
 				// 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 += string.Join(" ", nextNgram.Split(' ').Skip(1));
 				lastNgram = nextNgram;
 			}
 			wrandom.ClearContents();