//encoding CRF model from training corpus
public bool Learn(EncoderArgs args)
{
if (args.min_diff <= 0.0)
{
Console.WriteLine("eta must be > 0.0");
return(false);
}
if (args.C < 0.0)
{
Console.WriteLine("C must be >= 0.0");
return(false);
}
if (args.threads_num <= 0)
{
Console.WriteLine("thread must be > 0");
return(false);
}
if (args.hugeLexMemLoad > 0)
{
Console.WriteLine("Build feature lexical dictionary in huge mode[shrink when mem used rate:{0}%]", args.hugeLexMemLoad);
}
Console.WriteLine("Open and check training corpus and templates...");
ModelWritter modelWritter = new ModelWritter(args.threads_num, args.C, args.hugeLexMemLoad);
if (modelWritter.Open(args.strTemplateFileName, args.strTrainingCorpus) == false)
{
Console.WriteLine("Open training corpus or template file failed.");
return(false);
}
Console.WriteLine("Load training data and generate lexical features: ");
EncoderTagger[] xList = modelWritter.ReadAllRecords();
Console.WriteLine();
Console.WriteLine("Shrinking feature set [frequency is less than {0}]...", args.min_feature_freq);
modelWritter.Shrink(xList, args.min_feature_freq);
Console.Write("Saving model meta data...");
if (!modelWritter.SaveModelMetaData(args.strEncodedModelFileName))
{
Console.WriteLine("Failed!");
return(false);
}
else
{
Console.WriteLine("Success");
}
Console.WriteLine("Indexing feature set with {0} maximum slot usage rate threshold...", args.slot_usage_rate_threshold);
if (!modelWritter.BuildFeatureSetIntoIndex(args.strEncodedModelFileName, args.slot_usage_rate_threshold, args.debugLevel, args.strRetrainModelFileName))
{
Console.WriteLine("Failed!");
return(false);
}
else
{
Console.WriteLine("Success");
}
Console.WriteLine("Number of sentences: " + xList.Length);
Console.WriteLine("Number of features: " + modelWritter.feature_size());
Console.WriteLine("Number of thread(s): " + args.threads_num);
Console.WriteLine("Regularization type: " + args.regType.ToString());
Console.WriteLine("Freq: " + args.min_feature_freq);
Console.WriteLine("eta: " + args.min_diff);
Console.WriteLine("C: " + args.C);
bool orthant = false;
if (args.regType == REG_TYPE.L1)
{
orthant = true;
}
if (runCRF(xList, modelWritter, orthant, args) == false)
{
Console.WriteLine("Some warnings are raised during encoding...");
}
Console.Write("Saving model feature's weight...");
if (modelWritter.SaveFeatureWeight(args.strEncodedModelFileName) == false)
{
Console.WriteLine("Failed!");
return(false);
}
else
{
Console.WriteLine("Success.");
}
Console.WriteLine("\nDone!");
return(true);
}
//encoding CRF model from training corpus
public bool Learn(EncoderArgs args)
{
if (args.min_diff <= 0.0)
{
Console.WriteLine("eta must be > 0.0");
return false;
}
if (args.C < 0.0)
{
Console.WriteLine("C must be >= 0.0");
return false;
}
if (args.threads_num <= 0)
{
Console.WriteLine("thread must be > 0");
return false;
}
if (args.hugeLexMemLoad > 0)
{
Console.WriteLine("Build feature lexical dictionary in huge mode[shrink when mem used rate:{0}%]", args.hugeLexMemLoad);
}
Console.WriteLine("Open and check training corpus and templates...");
var modelWritter = new ModelWritter(args.threads_num, args.C, args.hugeLexMemLoad);
if (modelWritter.Open(args.strTemplateFileName, args.strTrainingCorpus) == false)
{
Console.WriteLine("Open training corpus or template file failed.");
return false;
}
Console.WriteLine("Load training data and generate lexical features: ");
var xList = modelWritter.ReadAllRecords();
Console.WriteLine();
Console.WriteLine("Shrinking feature set [frequency is less than {0}]...", args.min_feature_freq);
modelWritter.Shrink(xList, args.min_feature_freq);
Console.Write("Saving model meta data...");
if (!modelWritter.SaveModelMetaData(args.strEncodedModelFileName))
{
Console.WriteLine("Failed!");
return false;
}
else
{
Console.WriteLine("Success");
}
Console.WriteLine("Indexing feature set with {0} maximum slot usage rate threshold...", args.slot_usage_rate_threshold);
if (!modelWritter.BuildFeatureSetIntoIndex(args.strEncodedModelFileName, args.slot_usage_rate_threshold, args.debugLevel, args.strRetrainModelFileName))
{
Console.WriteLine("Failed!");
return false;
}
else
{
Console.WriteLine("Success");
}
Console.WriteLine("Number of sentences: " + xList.Length);
Console.WriteLine("Number of features: " + modelWritter.feature_size());
Console.WriteLine("Number of thread(s): " + args.threads_num);
Console.WriteLine("Regularization type: " + args.regType.ToString());
Console.WriteLine("Freq: " + args.min_feature_freq);
Console.WriteLine("eta: " + args.min_diff);
Console.WriteLine("C: " + args.C);
var orthant = false;
if (args.regType == REG_TYPE.L1)
{
orthant = true;
}
if (runCRF(xList, modelWritter, orthant, args) == false)
{
Console.WriteLine("Some warnings are raised during encoding...");
}
Console.Write("Saving model feature's weight...");
if (modelWritter.SaveFeatureWeight(args.strEncodedModelFileName) == false)
{
Console.WriteLine("Failed!");
return false;
}
else
{
Console.WriteLine("Success.");
}
Console.WriteLine("\nDone!");
return true;
}
bool runCRF(EncoderTagger[] x, ModelWritter modelWritter, bool orthant, EncoderArgs args)
{
double old_obj = double.MaxValue;
int converge = 0;
LBFGS lbfgs = new LBFGS(args.threads_num);
lbfgs.expected = new double[modelWritter.feature_size() + 1];
List <CRFEncoderThread> processList = new List <CRFEncoderThread>();
#if NO_SUPPORT_PARALLEL_LIB
#else
ParallelOptions parallelOption = new ParallelOptions();
parallelOption.MaxDegreeOfParallelism = args.threads_num;
#endif
//Initialize encoding threads
for (int i = 0; i < args.threads_num; i++)
{
CRFEncoderThread thread = new CRFEncoderThread();
thread.start_i = i;
thread.thread_num = args.threads_num;
thread.x = x;
thread.lbfgs = lbfgs;
thread.Init();
processList.Add(thread);
}
//Statistic term and result tags frequency
int termNum = 0;
int[] yfreq;
yfreq = new int[modelWritter.y_.Count];
foreach (EncoderTagger tagger in x)
{
termNum += tagger.word_num;
for (int j = 0; j < tagger.word_num; j++)
{
yfreq[tagger.answer_[j]]++;
}
}
//Iterative training
DateTime startDT = DateTime.Now;
double dMinErrRecord = 1.0;
for (int itr = 0; itr < args.max_iter; ++itr)
{
//Clear result container
lbfgs.obj = 0.0f;
lbfgs.err = 0;
lbfgs.zeroone = 0;
Array.Clear(lbfgs.expected, 0, lbfgs.expected.Length);
List <Thread> threadList = new List <Thread>();
for (int i = 0; i < args.threads_num; i++)
{
Thread thread = new Thread(new ThreadStart(processList[i].Run));
thread.Start();
threadList.Add(thread);
}
int[,] merr;
merr = new int[modelWritter.y_.Count, modelWritter.y_.Count];
for (int i = 0; i < args.threads_num; ++i)
{
threadList[i].Join();
lbfgs.obj += processList[i].obj;
lbfgs.err += processList[i].err;
lbfgs.zeroone += processList[i].zeroone;
//Calculate error
for (int j = 0; j < modelWritter.y_.Count; j++)
{
for (int k = 0; k < modelWritter.y_.Count; k++)
{
merr[j, k] += processList[i].merr[j, k];
}
}
}
long num_nonzero = 0;
long fsize = modelWritter.feature_size();
double[] alpha = modelWritter.alpha_;
if (orthant == true)
{
//L1 regularization
#if NO_SUPPORT_PARALLEL_LIB
for (long k = 1; k < fsize + 1; k++)
{
lbfgs.obj += Math.Abs(alpha[k] / modelWritter.cost_factor_);
if (alpha[k] != 0.0)
{
num_nonzero++;
}
}
#else
Parallel.For <double>(1, fsize + 1, parallelOption, () => 0, (k, loop, subtotal) =>
{
subtotal += Math.Abs(alpha[k] / modelWritter.cost_factor_);
if (alpha[k] != 0.0)
{
Interlocked.Increment(ref num_nonzero);
}
return(subtotal);
},
(subtotal) => // lock free accumulator
{
double initialValue;
double newValue;
do
{
initialValue = lbfgs.obj; // read current value
newValue = initialValue + subtotal; //calculate new value
}while (initialValue != Interlocked.CompareExchange(ref lbfgs.obj, newValue, initialValue));
}
);
#endif
}
else
{
//L2 regularization
num_nonzero = fsize;
#if NO_SUPPORT_PARALLEL_LIB
for (long k = 1; k < fsize + 1; k++)
{
lbfgs.obj += (alpha[k] * alpha[k] / (2.0 * modelWritter.cost_factor_));
lbfgs.expected[k] += (alpha[k] / modelWritter.cost_factor_);
}
#else
Parallel.For <double>(1, fsize + 1, parallelOption, () => 0, (k, loop, subtotal) =>
{
subtotal += (alpha[k] * alpha[k] / (2.0 * modelWritter.cost_factor_));
lbfgs.expected[k] += (alpha[k] / modelWritter.cost_factor_);
return(subtotal);
},
(subtotal) => // lock free accumulator
{
double initialValue;
double newValue;
do
{
initialValue = lbfgs.obj; // read current value
newValue = initialValue + subtotal; //calculate new value
}while (initialValue != Interlocked.CompareExchange(ref lbfgs.obj, newValue, initialValue));
}
);
#endif
}
//Show each iteration result
double diff = (itr == 0 ? 1.0f : Math.Abs(old_obj - lbfgs.obj) / old_obj);
old_obj = lbfgs.obj;
ShowEvaluation(x.Length, modelWritter, lbfgs, termNum, itr, merr, yfreq, diff, startDT, num_nonzero, args);
if (diff < args.min_diff)
{
converge++;
}
else
{
converge = 0;
}
if (itr > args.max_iter || converge == 3)
{
break; // 3 is ad-hoc
}
if (args.debugLevel > 0 && (double)lbfgs.zeroone / (double)x.Length < dMinErrRecord)
{
ConsoleColor cc = Console.ForegroundColor;
Console.ForegroundColor = ConsoleColor.Red;
Console.Write("[Debug Mode] ");
Console.ForegroundColor = cc;
Console.Write("Saving intermediate feature weights at current directory...");
//Save current best feature weight into file
dMinErrRecord = (double)lbfgs.zeroone / (double)x.Length;
modelWritter.SaveFeatureWeight("feature_weight_tmp");
Console.WriteLine("Done.");
}
int iret;
iret = lbfgs.optimize(alpha, modelWritter.cost_factor_, orthant);
if (iret <= 0)
{
return(false);
}
}
return(true);
}
bool runCRF(EncoderTagger[] x, ModelWritter modelWritter, bool orthant, EncoderArgs args)
{
var old_obj = double.MaxValue;
var converge = 0;
var lbfgs = new LBFGS(args.threads_num);
lbfgs.expected = new double[modelWritter.feature_size() + 1];
var processList = new List<CRFEncoderThread>();
#if NO_SUPPORT_PARALLEL_LIB
#else
var parallelOption = new ParallelOptions();
parallelOption.MaxDegreeOfParallelism = args.threads_num;
#endif
//Initialize encoding threads
for (var i = 0; i < args.threads_num; i++)
{
var thread = new CRFEncoderThread();
thread.start_i = i;
thread.thread_num = args.threads_num;
thread.x = x;
thread.lbfgs = lbfgs;
thread.Init();
processList.Add(thread);
}
//Statistic term and result tags frequency
var termNum = 0;
int[] yfreq;
yfreq = new int[modelWritter.y_.Count];
for (int index = 0; index < x.Length; index++)
{
var tagger = x[index];
termNum += tagger.word_num;
for (var j = 0; j < tagger.word_num; j++)
{
yfreq[tagger.answer_[j]]++;
}
}
//Iterative training
var startDT = DateTime.Now;
var dMinErrRecord = 1.0;
for (var itr = 0; itr < args.max_iter; ++itr)
{
//Clear result container
lbfgs.obj = 0.0f;
lbfgs.err = 0;
lbfgs.zeroone = 0;
Array.Clear(lbfgs.expected, 0, lbfgs.expected.Length);
var threadList = new List<Thread>();
for (var i = 0; i < args.threads_num; i++)
{
var thread = new Thread(processList[i].Run);
thread.Start();
threadList.Add(thread);
}
int[,] merr;
merr = new int[modelWritter.y_.Count, modelWritter.y_.Count];
for (var i = 0; i < args.threads_num; ++i)
{
threadList[i].Join();
lbfgs.obj += processList[i].obj;
lbfgs.err += processList[i].err;
lbfgs.zeroone += processList[i].zeroone;
//Calculate error
for (var j = 0; j < modelWritter.y_.Count; j++)
{
for (var k = 0; k < modelWritter.y_.Count; k++)
{
merr[j, k] += processList[i].merr[j, k];
}
}
}
long num_nonzero = 0;
var fsize = modelWritter.feature_size();
var alpha = modelWritter.alpha_;
if (orthant == true)
{
//L1 regularization
#if NO_SUPPORT_PARALLEL_LIB
for (long k = 1; k < fsize + 1; k++)
{
lbfgs.obj += Math.Abs(alpha[k] / modelWritter.cost_factor_);
if (alpha[k] != 0.0)
{
num_nonzero++;
}
}
#else
Parallel.For<double>(1, fsize + 1, parallelOption, () => 0, (k, loop, subtotal) =>
{
subtotal += Math.Abs(alpha[k] / modelWritter.cost_factor_);
if (alpha[k] != 0.0)
{
Interlocked.Increment(ref num_nonzero);
}
return subtotal;
},
(subtotal) => // lock free accumulator
{
double initialValue;
double newValue;
do
{
initialValue = lbfgs.obj; // read current value
newValue = initialValue + subtotal; //calculate new value
}
while (initialValue != Interlocked.CompareExchange(ref lbfgs.obj, newValue, initialValue));
}
);
#endif
}
else
{
//L2 regularization
num_nonzero = fsize;
#if NO_SUPPORT_PARALLEL_LIB
for (long k = 1; k < fsize + 1; k++)
{
lbfgs.obj += (alpha[k] * alpha[k] / (2.0 * modelWritter.cost_factor_));
lbfgs.expected[k] += (alpha[k] / modelWritter.cost_factor_);
}
#else
Parallel.For<double>(1, fsize + 1, parallelOption, () => 0, (k, loop, subtotal) =>
{
subtotal += (alpha[k] * alpha[k] / (2.0 * modelWritter.cost_factor_));
lbfgs.expected[k] += (alpha[k] / modelWritter.cost_factor_);
return subtotal;
},
(subtotal) => // lock free accumulator
{
double initialValue;
double newValue;
do
{
initialValue = lbfgs.obj; // read current value
newValue = initialValue + subtotal; //calculate new value
}
while (initialValue != Interlocked.CompareExchange(ref lbfgs.obj, newValue, initialValue));
}
);
#endif
}
//Show each iteration result
var diff = (itr == 0 ? 1.0f : Math.Abs(old_obj - lbfgs.obj) / old_obj);
old_obj = lbfgs.obj;
ShowEvaluation(x.Length, modelWritter, lbfgs, termNum, itr, merr, yfreq, diff, startDT, num_nonzero, args);
if (diff < args.min_diff)
{
converge++;
}
else
{
converge = 0;
}
if (itr > args.max_iter || converge == 3)
{
break; // 3 is ad-hoc
}
if (args.debugLevel > 0 && (double)lbfgs.zeroone / (double)x.Length < dMinErrRecord)
{
var cc = Console.ForegroundColor;
Console.ForegroundColor = ConsoleColor.Red;
Console.Write("[Debug Mode] ");
Console.ForegroundColor = cc;
Console.Write("Saving intermediate feature weights at current directory...");
//Save current best feature weight into file
dMinErrRecord = (double)lbfgs.zeroone / (double)x.Length;
modelWritter.SaveFeatureWeight("feature_weight_tmp");
Console.WriteLine("Done.");
}
int iret;
iret = lbfgs.optimize(alpha, modelWritter.cost_factor_, orthant);
if (iret <= 0)
{
return false;
}
}
return true;
}