public void Setup(bool debug)
{
var readModel = new ReadModel(InputModelFile);
var temp = new ReadModel(string.Concat(InputModelFile, ".featuresToK"));
_weightVector = new WeightVector(temp.GetFeatureToKdDictionary());
foreach (var pair in readModel.ModelIterator())
{
_weightVector.Add(pair);
}
_tags = new Tags(_tagList);
_viterbiForGlobalLinearModel = new ViterbiForGlobalLinearModel(_weightVector, _tags);
// read input file in a class and per line iterator.
var inputData = new ReadInputData(InputTestFile);
var writeModel = new WriteModel(_outputTestFile);
foreach (var line in inputData.GetSentence())
{
List<string> debugList;
var outputTags = _viterbiForGlobalLinearModel.Decode(line, debug, out debugList);
if (debug)
{
writeModel.WriteDataWithTagDebug(line, outputTags, debugList);
}
else
{
writeModel.WriteDataWithTag(line, outputTags);
}
}
writeModel.Flush();
}
public ComputeGradient(List<List<string>> inputSentence, List<List<string>> tagsList,
List<string> tagList, double lambda, double learningParam, FeatureCache cache, WriteModel logger)
{
Logger = logger;
_inputSentence = inputSentence;
_outputTagsList = tagsList;
_tagList = tagList;
_lambda = lambda;
_learningParam = learningParam;
_cache = cache;
forwardBackwordAlgos = new List<ForwardBackwordAlgo>();
_weightVector = null;
_twoGramsList = new string[4];
_twoGramPair = new KeyValuePair<string, string>[4];
var ngramTags = new Tags(_tagList);
int index = 0;
foreach (var ngram in ngramTags.GetNGramTags(2))
{
if (index >= _twoGramsList.Length)
{
Array.Resize(ref _twoGramsList, index+1);
Array.Resize(ref _twoGramPair, index + 1);
}
string[] split = ngram.Split(new[] { ':' });
_twoGramsList[index] = split[0] +"@#"+ split[1];
_twoGramPair[index] = new KeyValuePair<string, string>(split[0], split[1]);
index++;
}
}
public string Evalulate(string keyFile, string devFile, string dumpFile)
{
var keyModel = new ReadModel(keyFile);
var devModel = new ReadModel(devFile);
var dumpOutputModel = new WriteModel(dumpFile);
var keyIter = keyModel.ModelIterator().GetEnumerator();
var devIter = devModel.ModelIterator().GetEnumerator();
float expected = 0;
float correct = 0;
float found = 0;
float line = 0;
string dump;
while (keyIter.MoveNext() && devIter.MoveNext())
{
var key = keyIter.Current;
var dev = devIter.Current;
line++;
if (!key.Key.Equals(dev.Key))
{
dump = "line: " + line + " " + key.Key + " doesn't match " + dev.Key+"\r\n";
dumpOutputModel.WriteLine(dump);
dumpOutputModel.Flush();
return dump;
}
if (key.Value.Contains("LOCATION"))
{
expected++;
if (!dev.Value.Contains("LOCATION")) continue;
found++;
correct++;
}
else if (dev.Value.Contains("LOCATION"))
{
found++;
}
}
dump = "found: " + found + " expected: " + expected + " correct: " + correct +"\r\n";
dumpOutputModel.WriteLine(dump);
float precision = correct/found;
float recall = correct/expected;
float f1Score = (2*precision*recall)/(precision + recall);
dump += "precision\t recall \t f1score\t\r\n";
dumpOutputModel.WriteLine("precision\t recall \t f1score\t");
dump += precision.ToString(CultureInfo.InvariantCulture) + "\t" +
recall.ToString(CultureInfo.InvariantCulture) + "\t" +
f1Score.ToString(CultureInfo.InvariantCulture) +"\r\n";
dumpOutputModel.WriteLine(precision.ToString(CultureInfo.InvariantCulture)+"\t"+
recall.ToString(CultureInfo.InvariantCulture)+ "\t" +
f1Score.ToString(CultureInfo.InvariantCulture));
dumpOutputModel.Flush();
return dump;
}
public MapFeaturesToK(string outputFile, List<string> tagList)
{
_writeModel = new WriteModel(outputFile);
_tagList = tagList;
DictFeaturesToK = new Dictionary<string, int>();
DictKToFeatures = new Dictionary<int, string>();
FeatureCount = 0;
_tags = new Tags(tagList);
}
public void Dump()
{
var output = new WriteModel(string.Concat(_outputFile, ".preceptron"));
for (int index = 0; index < WeightVector.FeatureCount; index++)
{
output.WriteLine(string.Format("{0} {1} {2}", index,
MapFeatures.DictKToFeatures[index],
_useAvg ? AvgWeightVector.WeightArray[index] : WeightVector.WeightArray[index]));
}
output.Flush();
}
public void Dump(string outputFile, Dictionary<int, string> dictKtoFeature)
{
Console.WriteLine(DateTime.Now+" training is complete");
var output = new WriteModel(outputFile);
var sortedDictionary = from pair in _weightVector.WDictionary
orderby Math.Abs(pair.Value) descending
select pair;
foreach (var weight in sortedDictionary)
{
output.WriteLine(string.Format("{0} {1} {2}", weight.Key,
dictKtoFeature[weight.Key], weight.Value));
}
output.Flush();
}
public void Dump()
{
var output = new WriteModel(string.Concat(_outputFile, ".temp"));
var sortedDictionary = from pair in WeightVector.WDictionary
orderby Math.Abs(pair.Value) descending
select pair;
foreach (var weight in sortedDictionary)
{
output.WriteLine(string.Format("{0} {1}",
MapFeatures.DictKToFeatures[weight.Key], weight.Value));
//output.WriteLine(string.Format("{0} {1} {2}", weight.Key,
// MapFeatures.DictKToFeatures[weight.Key], weight.Value));
}
output.Flush();
}
public void Setup(bool debug)
{
Init();
// read input file in a class and per line iterator.
var inputData = new ReadInputData(InputTestFile);
var writeModel = new WriteModel(_outputTestFile);
foreach (var line in inputData.GetSentence())
{
List<string> debugList;
var outputTags = ViterbiForGLM.DecodeNew(line, debug, out debugList);
if (debug)
{
writeModel.WriteDataWithTagDebug(line, outputTags, debugList);
}
else
{
writeModel.WriteDataWithTag(line, outputTags);
}
}
writeModel.Flush();
}
internal static void CreateInputForCRF(string input, string output)
{
var reader = new ReadModel(input);
var keyWriter = new WriteModel(string.Concat(output, ".key"));
var devWriter = new WriteModel(string.Concat(output, ".key.dev"));
foreach (var line in reader.GetNextLine())
{
var words = line.Split(new[] {' '});
if (words.Length < 4)
continue;
foreach (var word in words)
{
if (string.IsNullOrEmpty(word.Trim()))
continue;
if (word.EndsWith("{LOCATION}"))
{
keyWriter.WriteLine(word.Replace("{LOCATION}", "") + " " + "LOCATION");
devWriter.WriteLine(word.Replace("{LOCATION}", ""));
}
else if (word.EndsWith("{LOCATION}."))
{
keyWriter.WriteLine(word.Replace("{LOCATION}.", ".") + " " + "LOCATION");
devWriter.WriteLine(word.Replace("{LOCATION}.", "."));
}
else
{
keyWriter.WriteLine(word + " " + "OTHER");
devWriter.WriteLine(word);
}
}
keyWriter.WriteLine("");
devWriter.WriteLine("");
}
keyWriter.Flush();
devWriter.Flush();
}
public ProcessRawText(string input, string output)
{
_reader = new ReadModel(input);
_writer = new WriteModel(output);
}
public void Parse(string input, string output)
{
var readModel= new ReadModel(input);
var writeModel = new WriteModel(output);
var writeDevModel = new WriteModel(output+".dev");
//var tempWrite = new WriteModel(output + "tempWrite");
var temp = new List<string>();
foreach (var line in readModel.GetNextLine())
{
var newLine = RemoveTags(line);
newLine = ReplaceTags(newLine);
newLine = RemoveAllTags(newLine);
if (string.IsNullOrEmpty(newLine)) continue;
//tempWrite.WriteLine(newLine);
var split = newLine.Split(new char[] {' '});
temp.AddRange(split.ToList());
//temp.Add("##NEWLINE##");
}
//tempWrite.Flush();
bool location = false;
var lastStr = string.Empty;
foreach (var tempStr in temp)
{
var str = tempStr.Trim();
if (string.IsNullOrEmpty(str))
{
lastStr = "";
continue;
}
//if (str.Equals("##NEWLINE##"))
{
if (!location && lastStr.EndsWith(".") && !IsSalutationAbbr(lastStr))
{
lastStr = string.Empty;
writeModel.WriteLine("");
writeDevModel.WriteLine("");
continue;
}
}
if (location)
{
if (str.Equals("##ENDTAG##"))
{
location = false;
lastStr = "";
continue;
}
writeModel.WriteLine(str + " " + "LOCATION");
writeDevModel.WriteLine(str);
lastStr = str;
continue;
}
if (str.Equals("##LOCATIONSTARTTAG##"))
{
lastStr = "";
location = true;
continue;
}
if (str.Equals("##ENDTAG##"))
{
lastStr = "";
continue;
}
writeModel.WriteLine(str + " "+ "OTHER");
writeDevModel.WriteLine(str);
lastStr = str;
}
writeModel.Flush();
writeDevModel.Flush();
}
static void TrainingTest(List<string> tags)
{
//const string modelFile = "../../data/gene.key.model";
//const string input = "../../data/gene.key";
const string modelFile = "../../data/training/tag.model.trial1";
const string input = "../../data/training/NYT_19980403_parsed.key";
string LoggerFile = "../../Logs/Log_"+DateTime.Now.ToFileTime()+".txt";
const int threadCount = 1;
var perceptron = new Perceptron(input, modelFile, tags);
perceptron.Train();
perceptron.ReMapFeatureToK();
//perceptron.Dump();
perceptron.MapFeatures.Dump();
perceptron.ReadInputs();
var featureCache = new FeatureCache(perceptron.InputSentences, tags,
perceptron.MapFeatures.DictFeaturesToK);
featureCache.CreateCache();
var logger = new WriteModel(LoggerFile);
var gradient = new ComputeGradient(perceptron.InputSentences, perceptron.TagsList,
tags, .1, featureCache, logger);
//perceptron.WeightVector.ResetAllToZero();
gradient.RunIterations(perceptron.WeightVector, 10, threadCount);
gradient.Dump(modelFile, perceptron.MapFeatures.DictKToFeatures);
}
public void Dump(string outputFile, Dictionary<int, string> dictKtoFeature)
{
Console.WriteLine(DateTime.Now+" training is complete");
var output = new WriteModel(outputFile);
for (int index = 0; index <_weightVector.FeatureCount; index++)
{
output.WriteLine(string.Format("{0} {1} {2}", index,
dictKtoFeature[index], _weightVector.WeightArray[index]));
}
//var sortedDictionary = from pair in _weightVector.WDictionary
// orderby Math.Abs(pair.Value) descending
// select pair;
//foreach (var weight in sortedDictionary)
//{
// output.WriteLine(string.Format("{0} {1} {2}", weight.Key,
// dictKtoFeature[weight.Key], weight.Value));
//}
output.Flush();
//Logger.Flush();
}
