文档部分
MQL5参考矩阵和向量方法机器学习ClassificationMetric 

ClassificationMetric

计算分类指标来评估预测数据相对于真实数据的质量。该方法适用于预测值的向量。

vector vector::ClassificationMetric(
   const vector&              vect_true,     // 真值向量
   ENUM_CLASSIFICATION_METRIC metric         // 指标类型
   );
 
 
vector vector::ClassificationMetric(
   const vector&              vect_true,     // 真值向量
   ENUM_CLASSIFICATION_METRIC metric         // 指标类型
   ENUM_AVERAGE_MODE          mode           // 平均模式
   );

参数

vect_true

[in] 真值向量。

metric

[in] ENUM_CLASSIFICATION_METRIC枚举的指标类型。应用CLASSIFICATION_TOP_K_ACCURACY， CLASSIFICATION_AVERAGE_PRECISION和CLASSIFICATION_ROC_AUC（在ClassificationScore方法中使用）以外的值。

mode

[in] ENUM_AVERAGE_MODE枚举的平均模式。用于CLASSIFICATION_F1，CLASSIFICATION_JACCARD， CLASSIFICATION_PRECISION和CLASSIFICATION_RECALL指标。

返回值

包含计算指标的向量。在AVERAGE_NONE平均模式的情况下，向量包含每个类别的指标值，而不进行平均。（例如，在二元分类的情况下，这将是分别用于'false'和'true'的两个指标）。

关于平均模式的说明

AVERAGE_BINARY只对二元分类有意义。

AVERAGE_MICRO ― 通过计算真阳性、假阴性和假阳性总数来计算全局指标。

AVERAGE_MACRO ― 计算每个标签的指标并找到其未加权平均值。这没有考虑标签不平衡。

AVERAGE_WEIGHTED ― 计算每个标签的指标并找到其按支持度加权的平均值（每个标签的真实实例数）。这改变了‘macro’以解决标签不平衡的问题；它可能会导致不在精确率和召回率之间的F分数。

示例：

   vector y_true={7,2,1,0,4,1,4,9,5,9,0,6,9,0,1,5,9,7,3,4,8,4,2,7,6,8,4,2,3,6};
   vector y_pred={7,2,1,0,4,1,4,9,5,9,0,6,9,0,1,5,9,7,3,4,2,9,4,9,5,9,2,7,7,0};
 
   vector accuracy=y_pred.ClassificationMetric(y_true,CLASSIFICATION_ACCURACY);
   Print("accuracy=",accuracy);
   vector balanced=y_pred.ClassificationMetric(y_true,CLASSIFICATION_BALANCED_ACCURACY);
   Print("balanced=",balanced);
   Print("");
 
   vector f1_micro=y_pred.ClassificationMetric(y_true,CLASSIFICATION_F1,AVERAGE_MICRO);
   Print("f1_micro=",f1_micro);
   vector f1_macro=y_pred.ClassificationMetric(y_true,CLASSIFICATION_F1,AVERAGE_MACRO);
   Print("f1_macro=",f1_macro);
   vector f1_weighted=y_pred.ClassificationMetric(y_true,CLASSIFICATION_F1,AVERAGE_WEIGHTED);
   Print("f1_weighted=",f1_weighted);
   vector f1_none=y_pred.ClassificationMetric(y_true,CLASSIFICATION_F1,AVERAGE_NONE);
   Print("f1_none=",f1_none);
   Print("");
 
   vector jaccard_micro=y_pred.ClassificationMetric(y_true,CLASSIFICATION_JACCARD,AVERAGE_MICRO);
   Print("jaccard_micro=",jaccard_micro);
   vector jaccard_macro=y_pred.ClassificationMetric(y_true,CLASSIFICATION_JACCARD,AVERAGE_MACRO);
   Print("jaccard_macro=",jaccard_macro);
   vector jaccard_weighted=y_pred.ClassificationMetric(y_true,CLASSIFICATION_JACCARD,AVERAGE_WEIGHTED);
   Print("jaccard_weighted=",jaccard_weighted);
   vector jaccard_none=y_pred.ClassificationMetric(y_true,CLASSIFICATION_JACCARD,AVERAGE_NONE);
   Print("jaccard_none=",jaccard_none);
   Print("");
 
   vector precision_micro=y_pred.ClassificationMetric(y_true,CLASSIFICATION_PRECISION,AVERAGE_MICRO);
   Print("precision_micro=",precision_micro);
   vector precision_macro=y_pred.ClassificationMetric(y_true,CLASSIFICATION_PRECISION,AVERAGE_MACRO);
   Print("precision_macro=",precision_macro);
   vector precision_weighted=y_pred.ClassificationMetric(y_true,CLASSIFICATION_PRECISION,AVERAGE_WEIGHTED);
   Print("precision_weighted=",precision_weighted);
   vector precision_none=y_pred.ClassificationMetric(y_true,CLASSIFICATION_PRECISION,AVERAGE_NONE);
   Print("precision_none=",precision_none);
   Print("");
 
   vector recall_micro=y_pred.ClassificationMetric(y_true,CLASSIFICATION_RECALL,AVERAGE_MICRO);
   Print("recall_micro=",recall_micro);
   vector recall_macro=y_pred.ClassificationMetric(y_true,CLASSIFICATION_RECALL,AVERAGE_MACRO);
   Print("recall_macro=",recall_macro);
   vector recall_weighted=y_pred.ClassificationMetric(y_true,CLASSIFICATION_RECALL,AVERAGE_WEIGHTED);
   Print("recall_weighted=",recall_weighted);
   vector recall_none=y_pred.ClassificationMetric(y_true,CLASSIFICATION_RECALL,AVERAGE_NONE);
   Print("recall_none=",recall_none);
   Print("");
 
//--- 二元分类
   vector y_pred_bin={0,1,0,1,1,0,0,0,1};
   vector y_true_bin={1,0,0,0,1,0,1,1,1};
 
   vector f1_bin=y_pred_bin.ClassificationMetric(y_true_bin,CLASSIFICATION_F1,AVERAGE_BINARY);
   Print("f1_bin=",f1_bin);
   vector jaccard_bin=y_pred_bin.ClassificationMetric(y_true_bin,CLASSIFICATION_JACCARD,AVERAGE_BINARY);
   Print("jaccard_bin=",jaccard_bin);
   vector precision_bin=y_pred_bin.ClassificationMetric(y_true_bin,CLASSIFICATION_PRECISION,AVERAGE_BINARY);
   Print("precision_bin=",precision_bin);
   vector recall_bin=y_pred_bin.ClassificationMetric(y_true_bin,CLASSIFICATION_RECALL,AVERAGE_BINARY);
   Print("recall_bin=",recall_bin);
 
 
/*
  accuracy=[0.6666666666666666]
  balanced=[0.6433333333333333]
  
  f1_micro=[0.6666666666666666]
  f1_macro=[0.6122510822510823]
  f1_weighted=[0.632049062049062]
  f1_none=[0.8571428571428571,1,0.3333333333333333,0.6666666666666666,0.6666666666666665,0.8,0.5,0.5714285714285715,0,0.7272727272727273]
  
  jaccard_micro=[0.5]
  jaccard_macro=[0.4921428571428572]
  jaccard_weighted=[0.5056349206349205]
  jaccard_none=[0.75,1,0.2,0.5,0.5,0.6666666666666666,0.3333333333333333,0.4,0,0.5714285714285714]
  
  precision_micro=[0.6666666666666666]
  precision_macro=[0.6571428571428571]
  precision_weighted=[0.6706349206349207]
  precision_none=[0.75,1,0.3333333333333333,1,0.75,0.6666666666666666,1,0.5,0,0.5714285714285714]
  
  recall_micro=[0.6666666666666666]
  recall_macro=[0.6433333333333333]
  recall_weighted=[0.6666666666666666]
  recall_none=[1,1,0.3333333333333333,0.5,0.6,1,0.3333333333333333,0.6666666666666666,0,1]
  
  f1_bin=[0.4444444444444445]
  jaccard_bin=[0.2857142857142857]
  precision_bin=[0.5]
  recall_bin=[0.4]
*/