public void LoadModels(BinarySerializer taggerModelSer, BinarySerializer lemmatizerModelSer)
{
Utils.ThrowException(taggerModelSer == null ? new ArgumentNullException("taggerModelSer") : null);
mLogger.Debug("Load", "Nalagam model za označevanje ...");
mSuffixTree = new PatriciaTree(taggerModelSer);
mFeatureSpace = Utils.LoadDictionary <string, int>(taggerModelSer);
mModel = new MaximumEntropyClassifierFast <string>(taggerModelSer);
if (lemmatizerModelSer != null)
{
mLogger.Debug("Load", "Nalagam model za lematizacijo ...");
mConsiderTags = lemmatizerModelSer.ReadBool();
mLemmatizer = new Lemmatizer(lemmatizerModelSer);
}
}
static void Main(string[] args)
{
try
{
if (args.Length < 2)
{
OutputHelp();
}
else
{
int cutOff = 2;
int numIter = 50;
int numThreads = 1;
string corpusFileName = null, modelFileName = null, lexiconFileName = null;
bool verbose = false;
if (ParseParams(args, ref verbose, ref cutOff, ref numIter, ref numThreads, ref corpusFileName, ref modelFileName, ref lexiconFileName))
{
Logger logger = Logger.GetRootLogger();
if (!verbose)
{
logger.LocalLevel = Logger.Level.Off;
logger.LocalProgressOutputType = Logger.ProgressOutputType.Off;
}
else
{
logger.LocalOutputType = Logger.OutputType.Custom;
Logger.CustomOutput = new Logger.CustomOutputDelegate(delegate(string loggerName, Logger.Level level, string funcName, Exception e,
string message, object[] msgArgs) { Console.WriteLine(message, msgArgs); });
}
Corpus corpus = new Corpus();
logger.Info(/*funcName=*/ null, "Nalagam učni korpus ...");
corpus.LoadFromXmlFile(corpusFileName, /*tagLen=*/ -1);
GC.Collect();
long oldMemUse = Process.GetCurrentProcess().PrivateMemorySize64;
PatriciaTree suffixTree = new PatriciaTree();
foreach (TaggedWord word in corpus.TaggedWords)
{
suffixTree.AddWordTagPair(word.WordLower, word.Tag);
}
if (lexiconFileName != null)
{
logger.Info(/*funcName=*/ null, "Nalagam leksikon ...");
StreamReader lexReader = new StreamReader(lexiconFileName);
string lexLine;
while ((lexLine = lexReader.ReadLine()) != null)
{
string[] lexData = lexLine.Split('\t');
suffixTree.AddWordTagPair(lexData[0].ToLower(), lexData[2]);
}
lexReader.Close();
}
GC.Collect();
long memUse = Process.GetCurrentProcess().PrivateMemorySize64;
Console.WriteLine("Poraba pomnilnika (drevo končnic): {0:0.00} MB", (double)(memUse - oldMemUse) / 1048576.0);
oldMemUse = memUse;
suffixTree.PropagateTags();
GC.Collect();
memUse = Process.GetCurrentProcess().PrivateMemorySize64;
Console.WriteLine("Poraba pomnilnika (propagirane oznake): {0:0.00} MB", (double)(memUse - oldMemUse) / 1048576.0);
MaximumEntropyClassifierFast <string> model = new MaximumEntropyClassifierFast <string>();
LabeledDataset <string, BinaryVector> dataset = new LabeledDataset <string, BinaryVector>();
Dictionary <string, int> featureSpace = new Dictionary <string, int>();
logger.Info(/*funcName=*/ null, "Pripravljam vektorje značilk ...");
for (int i = 0; i < corpus.TaggedWords.Count; i++)
{
logger.ProgressFast(Logger.Level.Info, /*funcName=*/ null, "{0} / {1}", i + 1, corpus.TaggedWords.Count);
BinaryVector featureVector = corpus.GenerateFeatureVector(i, featureSpace, /*extendFeatureSpace=*/ true, suffixTree);
dataset.Add(corpus.TaggedWords[i].Tag, featureVector);
}
logger.Info(/*funcName=*/ null, "Gradim model ...");
DateTime startTime = DateTime.Now;
model.CutOff = cutOff;
model.NumThreads = numThreads;
model.NumIter = numIter;
model.Train(dataset);
TimeSpan span = DateTime.Now - startTime;
logger.Info(/*funcName=*/ null, "Trajanje gradnje modela: {0:00}:{1:00}:{2:00}.{3:000}.", span.Hours, span.Minutes, span.Seconds, span.Milliseconds);
logger.Info(/*funcName=*/ null, "Zapisujem model ...");
BinarySerializer writer = new BinarySerializer(modelFileName, FileMode.Create);
suffixTree.Save(writer);
Utils.SaveDictionary(featureSpace, writer);
model.Save(writer);
writer.Close();
logger.Info(/*funcName=*/ null, "Končano.");
}
}
}
catch (Exception exception)
{
Console.WriteLine();
Console.WriteLine("*** Nepričakovana napaka. Podrobnosti: {0}\r\n{1}", exception, exception.StackTrace);
}
}
