public static bool MakeModel(string corpusPath, string modelSavePath)
{
var posSet = new SortedSet <string>(StrComparer.Default);
var dm = new DictionaryMaker();
for (int i = 0; i < 2; i++) //! 执行两遍，无非是将label的频率从1增加到2
{
foreach (var s in CoNLLUtil.LoadSentences(corpusPath))
{
foreach (var w in s.word)
{
AddPair(w.NAME, w.HEAD.NAME, w.DEPREL, dm);
AddPair(w.NAME, WrapTag(w.HEAD.POSTAG), w.DEPREL, dm);
AddPair(WrapTag(w.POSTAG), w.HEAD.NAME, w.DEPREL, dm);
AddPair(WrapTag(w.POSTAG), WrapTag(w.HEAD.POSTAG), w.DEPREL, dm);
posSet.Add(w.POSTAG);
}
}
}
var sb = new StringBuilder();
foreach (var pos in posSet)
{
sb.Append("cases \"" + pos + "\":\n");
}
File.WriteAllText(Config.Word_Nat_Weight_Model_Path, sb.ToString());
return(dm.SaveTxtTo(modelSavePath));
}
public CoNLLWord(int id, string lemma, string cpostag, string postag)
{
ID = id;
LEMMA = lemma;
CPOSTAG = cpostag;
POSTAG = postag;
NAME = CoNLLUtil.Compile(postag, lemma);
}
public static void MakeModel(string corpusPath, string modelSavePath)
{
var sentences = CoNLLUtil.LoadSentences(corpusPath);
var lines = new List <string>(sentences.Count * 30);
var sb = new StringBuilder();
//int id = 1;
for (int k = 0; k < sentences.Count; k++)
{
var s = sentences[k];
var edges = s.GetEdgeArr(); // 获取一句话中各词之间的依存关系（边）
var word = s.GetWordArrWithRoot(); // 获取一句话中的所有单词，包括虚根节点
var size = edges.GetLength(0);
for (int i = 0; i < size; i++)
{
for (int j = 0; j < size; j++)
{
if (i == j)
{
continue; // 词条与自身之间需要不考虑依存关系
}
var contexts = new List <string>();
// 从i出发到达j的边，可能存在这样的依存关系，也可能不存在，此时使用NULL代替
contexts.AddRange(GenerateSingleWordContext(word, i, "i"));
contexts.AddRange(GenerateSingleWordContext(word, j, "j"));
contexts.AddRange(GenerateUniContext(word, i, j));
foreach (var c in contexts)
{
sb.Append(c).Append(' ');
}
sb.Append(edges[i, j]);
lines.Add(sb.ToString());
sb.Clear();
}
}
}
File.WriteAllLines(modelSavePath, lines.ToArray());
}
