public TargetAnnotation SpaceTargetAnnotation()
{
TargetAnnotation result = new TargetAnnotation();
foreach (KeyValuePair <Label, double> labelAndReal in this.SpaceRealTargetLabelDic)
{
result.LabelAndTruthDic[labelAndReal.Key] = labelAndReal.Value >= 0;
}
return(result);
}
///// <summary>
///// 计算已知Source Annotation求Source Label为true的概率。
///// </summary>
///// <param name="sourceLabeltruth">Source Label和其Truth</param>
///// <returns>
///// Label的Truth与其在Annotation里的Truth相同：1。
///// Label的Truth与其在Annotation里的Truth不同：0。
///// </returns>
///// 废弃
//public double Pr_s_S(SourceLabeltruth sourceLabeltruth)
//{
// return this.LabelAndTruthDic[sourceLabeltruth.Label] == sourceLabeltruth.Truth ? 1 : 0;
//}
/// <summary>
/// 计算Pr(T|S)。
/// 两种遍历结果一致。
/// 论文公式10。
/// </summary>
/// <param name="T">作为条件的Target Annotation</param>
/// <returns>Pr(T|S)</returns>
public double Pr_T_S(TargetAnnotation T)
{
double result = 1;
#region T在里S在外（20141202公式第一行）。
/////先求P(T|s)，再求P(T|S)。
/////经验证，连加时（如PPT），谁在里谁在外都一样。根据乘法结合律。
/////连乘时又不一样了，每遍历一个s，P(t|T)就被乘一次。
//foreach (KeyValuePair<Label, bool> sourceLabelAndTruth in this.LabelAndTruthDic)//遍历s
//{
// SourceLabeltruth s = new SourceLabeltruth(sourceLabelAndTruth.Key, sourceLabelAndTruth.Value);
// double Pr_T_s = 1;//观察用，所以单独写出来。
// #region 非测试
// foreach (KeyValuePair<Label, bool> targetLabelAndTruth in T.LabelAndTruthDic)//遍历t
// {
// TargetLabeltruth t = new TargetLabeltruth(targetLabelAndTruth.Key, targetLabelAndTruth.Value);
// Pr_T_s *= t.Pr_T_t(T) * ProbabilityConstant.Pr_t_s[s][t];
// //Pr_T_s *= ProbabilityConstant.Pr_t_s[s][t];;
// }
// #endregion
// #region 测试
// //foreach(Label label in Constant.TargetTaxonomy.LabelArray)
// //{
// //TargetLabeltruth tTrue = new TargetLabeltruth(label, true);
// //double trueProbability = tTrue.Pr_T_t(T) * ProbabilityConstant.Pr_t_s[s][tTrue];
// //TargetLabeltruth tFalse = new TargetLabeltruth(label, false);
// //double falseProbability = tFalse.Pr_T_t(T) * ProbabilityConstant.Pr_t_s[s][tFalse];
// //}
// #endregion
// result *= Pr_T_s;
//}
#endregion
#region s在里t在外（20141202公式第二行）。先求P(t|S)，再求P(T|S)。
foreach (KeyValuePair <Label, bool> targetLabelAndTruth in T.LabelAndTruthDic)//先遍历t
{
#region P(t|S)未normalize
TargetLabeltruth t = new TargetLabeltruth(targetLabelAndTruth.Key, targetLabelAndTruth.Value);
double Pr_t_S = 1; //观察用，所以单独写出来。
foreach (KeyValuePair <Label, bool> sourceLabelAndTruth in this.LabelAndTruthDic) //遍历s
{
SourceLabeltruth s = new SourceLabeltruth(sourceLabelAndTruth.Key, sourceLabelAndTruth.Value);
Pr_t_S *= CascadedConstant.Pr_t_s[s][t];
}
//result *= t.Pr_T_t(T) * Pr_t_S;
result *= (Pr_t_S / Math.Pow(CascadedConstant.Pr_t[t], Constant.SourceTaxonomy.LabelArray.Length - 1));
#endregion
#region 测试,P(t|S)，normalize t
//TargetLabeltruth tTrue = new TargetLabeltruth(targetLabelAndTruth.Key, true);
//TargetLabeltruth tFalse = new TargetLabeltruth(targetLabelAndTruth.Key, false);
//double trueProbability = 1;
//double falseProbability = 1;
//foreach (KeyValuePair<Label, bool> sourceLabelAndTruth in this.LabelAndTruthDic)//遍历s
//{
// SourceLabeltruth s = new SourceLabeltruth(sourceLabelAndTruth.Key, sourceLabelAndTruth.Value);
// trueProbability *= ProbabilityConstant.Pr_t_s[s][tTrue];
// falseProbability *= ProbabilityConstant.Pr_t_s[s][tFalse];
//}
//if (targetLabelAndTruth.Value)
// result *= tTrue.Pr_T_t(T) * (trueProbability) / (trueProbability + falseProbability);
//else
// result *= tFalse.Pr_T_t(T) * (falseProbability) / (trueProbability + falseProbability);
#endregion
}
#endregion
return(result);
}