jellyfin/Emby.Common.Implementations/TextEncoding/NLangDetect/Detector.cs
2017-06-17 18:59:17 -04:00

372 lines
9.7 KiB
C#

using System;
using System.Collections.Generic;
using System.IO;
using System.Linq;
using System.Text;
using System.Text.RegularExpressions;
using NLangDetect.Core.Extensions;
using NLangDetect.Core.Utils;
namespace NLangDetect.Core
{
public class Detector
{
private const double _AlphaDefault = 0.5;
private const double _AlphaWidth = 0.05;
private const int _IterationLimit = 1000;
private const double _ProbThreshold = 0.1;
private const double _ConvThreshold = 0.99999;
private const int _BaseFreq = 10000;
private static readonly Regex _UrlRegex = new Regex("https?://[-_.?&~;+=/#0-9A-Za-z]+", RegexOptions.Compiled);
private static readonly Regex _MailRegex = new Regex("[-_.0-9A-Za-z]+@[-_0-9A-Za-z]+[-_.0-9A-Za-z]+", RegexOptions.Compiled);
private readonly Dictionary<string, ProbVector> _wordLangProbMap;
private readonly List<string> _langlist;
private StringBuilder _text;
private double[] _langprob;
private double _alpha = _AlphaDefault;
private const int _trialsCount = 7;
private int _maxTextLength = 10000;
private double[] _priorMap;
private int? _seed;
#region Constructor(s)
public Detector(DetectorFactory factory)
{
_wordLangProbMap = factory.WordLangProbMap;
_langlist = factory.Langlist;
_text = new StringBuilder();
_seed = factory.Seed;
}
#endregion
#region Public methods
public void SetAlpha(double alpha)
{
_alpha = alpha;
}
public void SetPriorMap(Dictionary<string, double> priorMap)
{
_priorMap = new double[_langlist.Count];
double sump = 0;
for (int i = 0; i < _priorMap.Length; i++)
{
string lang = _langlist[i];
if (priorMap.ContainsKey(lang))
{
double p = priorMap[lang];
if (p < 0)
{
throw new NLangDetectException("Prior probability must be non-negative.", ErrorCode.InitParamError);
}
_priorMap[i] = p;
sump += p;
}
}
if (sump <= 0)
{
throw new NLangDetectException("More one of prior probability must be non-zero.", ErrorCode.InitParamError);
}
for (int i = 0; i < _priorMap.Length; i++)
{
_priorMap[i] /= sump;
}
}
public void SetMaxTextLength(int max_text_length)
{
_maxTextLength = max_text_length;
}
// TODO IMM HI: TextReader?
public void Append(StreamReader streamReader)
{
var buf = new char[_maxTextLength / 2];
while (_text.Length < _maxTextLength && !streamReader.EndOfStream)
{
int length = streamReader.Read(buf, 0, buf.Length);
Append(new string(buf, 0, length));
}
}
public void Append(string text)
{
text = _UrlRegex.Replace(text, " ");
text = _MailRegex.Replace(text, " ");
char pre = '\0';
for (int i = 0; i < text.Length && i < _maxTextLength; i++)
{
char c = NGram.Normalize(text[i]);
if (c != ' ' || pre != ' ')
{
_text.Append(c);
}
pre = c;
}
}
private void CleanText()
{
int latinCount = 0, nonLatinCount = 0;
for (int i = 0; i < _text.Length; i++)
{
char c = _text[i];
if (c <= 'z' && c >= 'A')
{
latinCount++;
}
else if (c >= '\u0300' && c.GetUnicodeBlock() != UnicodeBlock.LatinExtendedAdditional)
{
nonLatinCount++;
}
}
if (latinCount * 2 < nonLatinCount)
{
var textWithoutLatin = new StringBuilder();
for (int i = 0; i < _text.Length; i++)
{
char c = _text[i];
if (c > 'z' || c < 'A')
{
textWithoutLatin.Append(c);
}
}
_text = textWithoutLatin;
}
}
public string Detect()
{
List<Language> probabilities = GetProbabilities();
return
probabilities.Count > 0
? probabilities[0].Name
: null;
}
public List<Language> GetProbabilities()
{
if (_langprob == null)
{
DetectBlock();
}
List<Language> list = SortProbability(_langprob);
return list;
}
#endregion
#region Private helper methods
private static double NormalizeProb(double[] probs)
{
double maxp = 0, sump = 0;
sump += probs.Sum();
for (int i = 0; i < probs.Length; i++)
{
double p = probs[i] / sump;
if (maxp < p)
{
maxp = p;
}
probs[i] = p;
}
return maxp;
}
private static string UnicodeEncode(string word)
{
var resultSb = new StringBuilder();
foreach (char ch in word)
{
if (ch >= '\u0080')
{
string st = string.Format("{0:x}", 0x10000 + ch);
while (st.Length < 4)
{
st = "0" + st;
}
resultSb
.Append("\\u")
.Append(st.SubSequence(1, 5));
}
else
{
resultSb.Append(ch);
}
}
return resultSb.ToString();
}
private void DetectBlock()
{
CleanText();
List<string> ngrams = ExtractNGrams();
if (ngrams.Count == 0)
{
throw new NLangDetectException("no features in text", ErrorCode.CantDetectError);
}
_langprob = new double[_langlist.Count];
Random rand = (_seed.HasValue ? new Random(_seed.Value) : new Random());
for (int t = 0; t < _trialsCount; t++)
{
double[] prob = InitProbability();
// TODO IMM HI: verify it works
double alpha = _alpha + rand.NextGaussian() * _AlphaWidth;
for (int i = 0; ; i++)
{
int r = rand.Next(ngrams.Count);
UpdateLangProb(prob, ngrams[r], alpha);
if (i % 5 == 0)
{
if (NormalizeProb(prob) > _ConvThreshold || i >= _IterationLimit)
{
break;
}
}
}
for (int j = 0; j < _langprob.Length; j++)
{
_langprob[j] += prob[j] / _trialsCount;
}
}
}
private double[] InitProbability()
{
var prob = new double[_langlist.Count];
if (_priorMap != null)
{
for (int i = 0; i < prob.Length; i++)
{
prob[i] = _priorMap[i];
}
}
else
{
for (int i = 0; i < prob.Length; i++)
{
prob[i] = 1.0 / _langlist.Count;
}
}
return prob;
}
private List<string> ExtractNGrams()
{
var list = new List<string>();
NGram ngram = new NGram();
for (int i = 0; i < _text.Length; i++)
{
ngram.AddChar(_text[i]);
for (int n = 1; n <= NGram.GramsCount; n++)
{
string w = ngram.Get(n);
if (w != null && _wordLangProbMap.ContainsKey(w))
{
list.Add(w);
}
}
}
return list;
}
private void UpdateLangProb(double[] prob, string word, double alpha)
{
if (word == null || !_wordLangProbMap.ContainsKey(word))
{
return;
}
ProbVector langProbMap = _wordLangProbMap[word];
double weight = alpha / _BaseFreq;
for (int i = 0; i < prob.Length; i++)
{
prob[i] *= weight + langProbMap[i];
}
}
private List<Language> SortProbability(double[] prob)
{
var list = new List<Language>();
for (int j = 0; j < prob.Length; j++)
{
double p = prob[j];
if (p > _ProbThreshold)
{
for (int i = 0; i <= list.Count; i++)
{
if (i == list.Count || list[i].Probability < p)
{
list.Insert(i, new Language(_langlist[j], p));
break;
}
}
}
}
return list;
}
#endregion
}
}