/// <summary>
/// Generate a probability hash from raw text data, and serialize
/// the hash for later use.
/// </summary>
/// <param name="topic"></param>
/// <param name="order"></param>
/// <returns></returns>
public static ProbabilityHash Parse(string topic, int order)
{
var probabilityHash = new ProbabilityHash(order);
// This was a rushed design decision. I actually change the working directory to
// get to the corpus data. A lock needs to be held because the client can be multithreaded,
// and a race condition could cause the directory to be entered twice.
lock (DataDirectories.DirectoryControl)
{
DataDirectories.EnterDirectory(RawTextWriter.DirectoryName);
// Find all subtopic files for the given topic
var filenames = FindFilenames(topic);
// If there are no subtopic files, then the topic must not
// exist on disk.
if (filenames.Count == 0)
{
DataDirectories.LeaveDirectory();
throw new FileNotFoundException();
}
// Read each subtopic into the probability hash
foreach (var subtopic in filenames.Keys)
{
var data = new RawTextParseData()
{
ProbabilityHash = probabilityHash,
PreviousWords = new List<string>(),
Subtopic = subtopic
};
using (var file = File.OpenText(filenames[subtopic]))
{
while (!file.EndOfStream)
{
data.Line = file.ReadLine();
ParseLine(data);
}
}
}
// Finalize the probability hash so it can be used
// in the Markov chain.
probabilityHash.Finalize();
DataDirectories.LeaveDirectory();
}
// Produce a serialized version of the probability hash for later use.
CorpusWriter.Write(topic, probabilityHash);
return probabilityHash;
}
/// <summary>
/// Generate Markov states and state transitions from a line of raw text data.
/// </summary>
/// <param name="data"></param>
private static void ParseLine(RawTextParseData data)
{
// Remove citation references from the line
string line = wikiReferencePattern.Replace(data.Line, "");
// Clean the words of other unwanted characters, such as double quotes.
var words = TextCorpus.GetCleanWords(line);
foreach (string word in words)
{
if (data.PreviousWords.Count < data.ProbabilityHash.Order)
{
// We first have to find the first state by populating the
// "PreviousWords" list
data.PreviousWords.Add(word);
}
else
{
// Once we have an initial state, we can start generating transition edges
// and new states based on each successive word
data.ProbabilityHash.Add(word, data.Subtopic, data.PreviousWords);
data.PreviousWords.Add(word);
data.PreviousWords.RemoveAt(0);
}
}
}
