public int optimize(double[] x, double C, bool orthant)
{
const long msize = 5;
var size = x.LongLength - 1;
if (w == null || w.LongLength == 0)
{
iflag_ = 0;
w = new FixedBigArray <double>(size * (2 * msize + 1) + 2 * msize, 1);
diag = new double[size + 1];
if (orthant == true)
{
xi = new double[size + 1];
v = new double[size + 1];
}
}
if (orthant == true)
{
pseudo_gradient(x, C);
}
else
{
v = expected;
}
lbfgs_optimize(msize, x, orthant, C);
if (iflag_ < 0)
{
Console.WriteLine("routine stops with unexpected error");
return(-1);
}
return(iflag_);
}
//Generate feature string and its id list
public void GenerateLexicalIdList(out IList <string> keyList, out IList <int> valList)
{
var fixArrayKey = new FixedBigArray <string>(Size, 0);
keyList = fixArrayKey;
var fixArrayValue = new FixedBigArray <int>(Size, 0);
valList = fixArrayValue;
Parallel.For(0, arrayFeatureFreqSize, parallelOption, i =>
{
fixArrayKey[i] = arrayFeatureFreq[i].strFeature;
fixArrayValue[i] = (int)(arrayFeatureFreq[i].value);
});
}
private double ddot_(long size, FixedBigArray <double> dx, long dx_idx, FixedBigArray <double> dy, long dy_idx)
{
double ret = 0.0f;
Parallel.For <double>(0, size, parallelOption, () => 0, (i, loop, subtotal) =>
{
subtotal += dx[i + dx_idx] * dy[i + dy_idx];
return(subtotal);
},
(subtotal) => // lock free accumulator
{
double initialValue;
double newValue;
do
{
initialValue = ret; // read current value
newValue = initialValue + subtotal; //calculate new value
}while (initialValue != Interlocked.CompareExchange(ref ret, newValue, initialValue));
});
return(ret);
}
//Generate feature string and its id list
public void GenerateLexicalIdList(out IList<string> keyList, out IList<int> valList)
{
var fixArrayKey = new FixedBigArray<string>(Size, 0);
keyList = fixArrayKey;
var fixArrayValue = new FixedBigArray<int>(Size, 0);
valList = fixArrayValue;
#if NO_SUPPORT_PARALLEL_LIB
for (long i = 0;i < arrayFeatureFreqSize;i++)
#else
Parallel.For(0, arrayFeatureFreqSize, parallelOption, i =>
#endif
{
fixArrayKey[i] = arrayFeatureFreq[i].strFeature;
fixArrayValue[i] = (int)(arrayFeatureFreq[i].value);
}
#if NO_SUPPORT_PARALLEL_LIB
#else
);
#endif
}
static void Main(string[] args)
{
FixedBigArray <int> ba = new FixedBigArray <int>(1024, 0);
ba[1] = 1;
}
public void Process(string strModelFileName, string strShrinkedModelFileName, int thread_num_ = 1)
{
var sr = new StreamReader(strModelFileName);
string strLine;
//读入版本号
strLine = sr.ReadLine();
var version = uint.Parse(strLine.Split(':')[1].Trim());
if (version == CRFSharp.Utils.MODEL_TYPE_SHRINKED)
{
Console.WriteLine("The input model has been shrinked");
return;
}
//读入cost_factor
strLine = sr.ReadLine();
var cost_factor_ = double.Parse(strLine.Split(':')[1].Trim());
//读入maxid
strLine = sr.ReadLine();
var maxid_ = long.Parse(strLine.Split(':')[1].Trim());
//读入xsize
strLine = sr.ReadLine();
var xsize_ = uint.Parse(strLine.Split(':')[1].Trim());
//读入空行
strLine = sr.ReadLine();
//读入待标注的标签
var y_ = new List<string>();
while (true)
{
strLine = sr.ReadLine();
if (strLine.Length == 0)
{
break;
}
y_.Add(strLine);
}
//读入unigram和bigram模板
var unigram_templs_ = new List<string>();
var bigram_templs_ = new List<string>();
while (sr.EndOfStream == false)
{
strLine = sr.ReadLine();
if (strLine.Length == 0)
{
break;
}
if (strLine[0] == 'U')
{
unigram_templs_.Add(strLine);
}
if (strLine[0] == 'B')
{
bigram_templs_.Add(strLine);
}
}
sr.Close();
//Load all features alpha data
var filename_alpha = strModelFileName + ".alpha";
var filename_shrink_alpha = strShrinkedModelFileName + ".alpha";
var sr_alpha = new StreamReader(filename_alpha);
var br_alpha = new BinaryReader(sr_alpha.BaseStream);
var sw_alpha = new StreamWriter(filename_shrink_alpha);
var bw_alpha = new BinaryWriter(sw_alpha.BaseStream);
long shrinked_alpha_size = 0;
//Only reserve non-zero feature weights and save them into file as two-tuples format
var alpha_ = new FixedBigArray<double>(maxid_ + 1, 0);
for (long i = 0; i < maxid_; i++)
{
alpha_[i] = br_alpha.ReadSingle();
if (alpha_[i] != 0)
{
bw_alpha.Write(i);
bw_alpha.Write((float)alpha_[i]);
shrinked_alpha_size++;
}
}
br_alpha.Close();
bw_alpha.Close();
//Only reserved lexical feature whose weights is non-zero
var varValue = new VarBigArray<int>(1024);
var varFeature = new VarBigArray<string>(1024);
var feaCnt = 0;
var filename_feature = strModelFileName + ".feature.raw_text";
var sr_fea = new StreamReader(filename_feature);
while (sr_fea.EndOfStream == false)
{
strLine = sr_fea.ReadLine();
var items = strLine.Split('\t');
var strFeature = items[0];
var key = int.Parse(items[1]);
var size = (strFeature[0] == 'U' ? y_.Count : y_.Count * y_.Count);
var hasAlpha = false;
for (var i = key; i < key + size; i++)
{
if (alpha_[i] != 0)
{
hasAlpha = true;
break;
}
}
if (hasAlpha == true)
{
varFeature[feaCnt] = strFeature;
varValue[feaCnt] = key;
feaCnt++;
}
}
sr_fea.Close();
Console.WriteLine("Shrink feature size from {0} to {1}", maxid_, shrinked_alpha_size);
maxid_ = shrinked_alpha_size;
//Build new lexical feature
var val = new FixedBigArray<int>(feaCnt, 0);
var fea = new FixedBigArray<string>(feaCnt, 0);
for (var i = 0; i < feaCnt; i++)
{
fea[i] = varFeature[i];
val[i] = varValue[i];
}
varFeature = null;
varValue = null;
var da = new DoubleArrayTrieBuilder(thread_num_);
if (da.build(fea, val, 0.95) == false)
{
Console.WriteLine("Build lexical dictionary failed.");
return;
}
da.save(strShrinkedModelFileName + ".feature");
var tofs = new StreamWriter(strShrinkedModelFileName);
// header
tofs.WriteLine("version: " + CRFSharp.Utils.MODEL_TYPE_SHRINKED);
tofs.WriteLine("cost-factor: " + cost_factor_);
tofs.WriteLine("maxid: " + maxid_);
tofs.WriteLine("xsize: " + xsize_);
tofs.WriteLine();
// y
for (var i = 0; i < y_.Count; ++i)
{
tofs.WriteLine(y_[i]);
}
tofs.WriteLine();
// template
for (var i = 0; i < unigram_templs_.Count; ++i)
{
tofs.WriteLine(unigram_templs_[i]);
}
for (var i = 0; i < bigram_templs_.Count; ++i)
{
tofs.WriteLine(bigram_templs_[i]);
}
tofs.Close();
}
public void mcsrch(double[] x, double f, double[] g, FixedBigArray <double> s, long s_idx,
ref double stp, ref long info, ref long nfev, double[] wa)
{
var size = x.LongLength - 1;
/* Parameter adjustments */
if (info == -1)
{
info = 0;
nfev++;
var dg = ddot_(size, g, 1, s, s_idx + 1);
var ftest1 = finit + stp * dgtest;
if (brackt && ((stp <= stmin || stp >= stmax) || infoc == 0))
{
info = 6;
Console.WriteLine("MCSRCH warning: Rounding errors prevent further progress.There may not be a step which satisfies the sufficient decrease and curvature conditions. Tolerances may be too small.");
Console.WriteLine("bracket: {0}, stp:{1}, stmin:{2}, stmax:{3}, infoc:{4}", brackt, stp, stmin, stmax, infoc);
}
if (stp == lb3_1_stpmax && f <= ftest1 && dg <= dgtest)
{
info = 5;
Console.WriteLine("MCSRCH warning: The step is too large.");
}
if (stp == lb3_1_stpmin && (f > ftest1 || dg >= dgtest))
{
info = 4;
Console.WriteLine("MCSRCH warning: The step is too small.");
Console.WriteLine("stp:{0}, lb3_1_stpmin:{1}, f:{2}, ftest1:{3}, dg:{4}, dgtest:{5}", stp, lb3_1_stpmin, f, ftest1, dg, dgtest);
}
if (nfev >= maxfev)
{
info = 3;
Console.WriteLine("MCSRCH warning: More than {0} function evaluations were required at the present iteration.", maxfev);
}
if (brackt && stmax - stmin <= xtol * stmax)
{
info = 2;
Console.WriteLine("MCSRCH warning: Relative width of the interval of uncertainty is at most xtol.");
}
if (f <= ftest1 && Math.Abs(dg) <= lb3_1_gtol * (-dginit))
{
info = 1;
}
if (info != 0)
{
return;
}
if (stage1 && f <= ftest1 && dg >= Math.Min(ftol, lb3_1_gtol) * dginit)
{
stage1 = false;
}
if (stage1 && f <= fx && f > ftest1)
{
var fm = f - stp * dgtest;
var fxm = fx - stx * dgtest;
var fym = fy - sty * dgtest;
var dgm = dg - dgtest;
var dgxm = dgx - dgtest;
var dgym = dgy - dgtest;
mcstep(ref stx, ref fxm, ref dgxm, ref sty, ref fym, ref dgym, ref stp, fm, dgm, ref brackt,
stmin, stmax, ref infoc);
fx = fxm + stx * dgtest;
fy = fym + sty * dgtest;
dgx = dgxm + dgtest;
dgy = dgym + dgtest;
}
else
{
mcstep(ref stx, ref fx, ref dgx, ref sty, ref fy, ref dgy, ref stp, f, dg, ref brackt,
stmin, stmax, ref infoc);
}
if (brackt)
{
var d1 = 0.0;
d1 = sty - stx;
if (Math.Abs(d1) >= p66 * width1)
{
stp = stx + p5 * (sty - stx);
}
width1 = width;
d1 = sty - stx;
width = Math.Abs(d1);
}
}
else
{
infoc = 1;
if (size <= 0 || stp <= 0.0)
{
return;
}
dginit = ddot_(size, g, 1, s, s_idx + 1);
if (dginit >= 0.0)
{
return;
}
brackt = false;
stage1 = true;
nfev = 0;
finit = f;
dgtest = ftol * dginit;
width = lb3_1_stpmax - lb3_1_stpmin;
width1 = width / p5;
Parallel.For(1, size + 1, parallelOption, i =>
{
wa[i] = x[i];
}
);
stx = 0.0;
fx = finit;
dgx = dginit;
sty = 0.0;
fy = finit;
dgy = dginit;
}
if (brackt)
{
stmin = Math.Min(stx, sty);
stmax = Math.Max(stx, sty);
}
else
{
stmin = stx;
stmax = stp + xtrapf * (stp - stx);
}
stp = Math.Max(stp, lb3_1_stpmin);
stp = Math.Min(stp, lb3_1_stpmax);
if ((brackt && ((stp <= stmin || stp >= stmax) ||
nfev >= maxfev - 1 || infoc == 0)) ||
(brackt && (stmax - stmin <= xtol * stmax)))
{
stp = stx;
}
var stp_t = stp;
Parallel.For(1, size + 1, parallelOption, i =>
{
x[i] = (wa[i] + stp_t * s[s_idx + i]);
});
info = -1;
}
public void Process(string strModelFileName, string strShrinkedModelFileName, int thread_num_ = 1)
{
var sr = new StreamReader(strModelFileName);
string strLine;
//读入版本号
strLine = sr.ReadLine();
var version = uint.Parse(strLine.Split(':')[1].Trim());
if (version == CRFSharp.Utils.MODEL_TYPE_SHRINKED)
{
Console.WriteLine("The input model has been shrinked");
return;
}
//读入cost_factor
strLine = sr.ReadLine();
var cost_factor_ = double.Parse(strLine.Split(':')[1].Trim());
//读入maxid
strLine = sr.ReadLine();
var maxid_ = long.Parse(strLine.Split(':')[1].Trim());
//读入xsize
strLine = sr.ReadLine();
var xsize_ = uint.Parse(strLine.Split(':')[1].Trim());
//读入空行
strLine = sr.ReadLine();
//读入待标注的标签
var y_ = new List <string>();
while (true)
{
strLine = sr.ReadLine();
if (strLine.Length == 0)
{
break;
}
y_.Add(strLine);
}
//读入unigram和bigram模板
var unigram_templs_ = new List <string>();
var bigram_templs_ = new List <string>();
while (sr.EndOfStream == false)
{
strLine = sr.ReadLine();
if (strLine.Length == 0)
{
break;
}
if (strLine[0] == 'U')
{
unigram_templs_.Add(strLine);
}
if (strLine[0] == 'B')
{
bigram_templs_.Add(strLine);
}
}
sr.Close();
//Load all features alpha data
var filename_alpha = strModelFileName + ".alpha";
var filename_shrink_alpha = strShrinkedModelFileName + ".alpha";
var sr_alpha = new StreamReader(filename_alpha);
var br_alpha = new BinaryReader(sr_alpha.BaseStream);
var sw_alpha = new StreamWriter(filename_shrink_alpha);
var bw_alpha = new BinaryWriter(sw_alpha.BaseStream);
long shrinked_alpha_size = 0;
//Only reserve non-zero feature weights and save them into file as two-tuples format
var alpha_ = new FixedBigArray <double>(maxid_ + 1, 0);
for (long i = 0; i < maxid_; i++)
{
alpha_[i] = br_alpha.ReadSingle();
if (alpha_[i] != 0)
{
bw_alpha.Write(i);
bw_alpha.Write((float)alpha_[i]);
shrinked_alpha_size++;
}
}
br_alpha.Close();
bw_alpha.Close();
//Only reserved lexical feature whose weights is non-zero
var varValue = new VarBigArray <int>(1024);
var varFeature = new VarBigArray <string>(1024);
var feaCnt = 0;
var filename_feature = strModelFileName + ".feature.raw_text";
var sr_fea = new StreamReader(filename_feature);
while (sr_fea.EndOfStream == false)
{
strLine = sr_fea.ReadLine();
var items = strLine.Split('\t');
var strFeature = items[0];
var key = int.Parse(items[1]);
var size = (strFeature[0] == 'U' ? y_.Count : y_.Count * y_.Count);
var hasAlpha = false;
for (var i = key; i < key + size; i++)
{
if (alpha_[i] != 0)
{
hasAlpha = true;
break;
}
}
if (hasAlpha == true)
{
varFeature[feaCnt] = strFeature;
varValue[feaCnt] = key;
feaCnt++;
}
}
sr_fea.Close();
Console.WriteLine("Shrink feature size from {0} to {1}", maxid_, shrinked_alpha_size);
maxid_ = shrinked_alpha_size;
//Build new lexical feature
var val = new FixedBigArray <int>(feaCnt, 0);
var fea = new FixedBigArray <string>(feaCnt, 0);
for (var i = 0; i < feaCnt; i++)
{
fea[i] = varFeature[i];
val[i] = varValue[i];
}
varFeature = null;
varValue = null;
var da = new DoubleArrayTrieBuilder(thread_num_);
if (da.build(fea, val, 0.95) == false)
{
Console.WriteLine("Build lexical dictionary failed.");
return;
}
da.save(strShrinkedModelFileName + ".feature");
var tofs = new StreamWriter(strShrinkedModelFileName);
// header
tofs.WriteLine("version: " + CRFSharp.Utils.MODEL_TYPE_SHRINKED);
tofs.WriteLine("cost-factor: " + cost_factor_);
tofs.WriteLine("maxid: " + maxid_);
tofs.WriteLine("xsize: " + xsize_);
tofs.WriteLine();
// y
for (var i = 0; i < y_.Count; ++i)
{
tofs.WriteLine(y_[i]);
}
tofs.WriteLine();
// template
for (var i = 0; i < unigram_templs_.Count; ++i)
{
tofs.WriteLine(unigram_templs_[i]);
}
for (var i = 0; i < bigram_templs_.Count; ++i)
{
tofs.WriteLine(bigram_templs_[i]);
}
tofs.Close();
}
static void Main(string[] args)
{
FixedBigArray<int> ba = new FixedBigArray<int>(1024, 0);
ba[1] = 1;
}