diff --git a/ml101/performance-metrics.md b/ml101/performance-metrics.md
index a13d870..e5bc769 100644
--- a/ml101/performance-metrics.md
+++ b/ml101/performance-metrics.md
@@ -9,14 +9,29 @@ In signal detection theory, a receiver operating characteristic (ROC), or simply
 Most binary classifiers give a prediction probability for positive and negative classes. If you set a threshold say, 0.6, you will get a Recall(TPR) and False alarm(FPR). then you vary this threshold value, you will get a group of points. threshold value = 0 corresponds to the point (1,0) while threshold value = 1 corresponds to point(0,0)
 
 ![](http://upload-images.jianshu.io/upload_images/145616-2063bb79c3684a8a.png?imageMogr2/auto-orient/strip%7CimageView2/2/w/1240)
+
 [1] [机器学习之分类性能度量指标 : ROC曲线、AUC值、正确率、召回率](http://www.jianshu.com/p/c61ae11cc5f6)
 
 
-## Why use ROC/AUC?
+## Why/When use ROC/AUC?
 
 The AUC value is equivalent to the probability that a randomly chosen positive example 
 i s ranked higher than a randomly chosen negative example.
-When data sets are imbalanced, ROC/AUC is more stable than Recall, F1, precision..
+When data sets are imbalanced, ROC/AUC is more stable than Recall, F1, precision.
+
+![stability_1](https://pic2.zhimg.com/74db397e36eabfb505abedd68f15bd57_r.jpg)
+
+To explain the stability of ROC/AUC over Precision/Recall, check the figure above. (a), (b) are the original curves, (c), (d) are the cureves when we increase the negative instances 10 times. It's obvious that the ROC/AUC doesn't change much, while Precision/Recall curse changes a lot.
+
+![stability_2](https://pic2.zhimg.com/v2-a97e84d7726cfe6989e513c3bad8ee79_r.jpg)
+
+With that being that, the stability of ROC/AUC over Precision/Recall Curve is actually a double-edged sword. Because that also means when tha data are imbalanced, you cannot tell the difference from ROC/AUC but from Precision/Recall. Check the figure above, from the ROC/AUC, both algorithms are nearly perfect, but when you check the Precision/Recall curve, it's easy to tell they are far away from being perfect. In fact, tons of improvements are needed for both of the algorithms.
+
+
+**Take-away:**
+- ROC/AUC is more stable than Precision/Recall. (double-edged sword)
+- when assume data are balanced, use ROC/AUC. 
+- when assume data are inbalanced, use Precision/Recall.
 
 
 ## Confusion matrix
@@ -26,4 +41,4 @@ $$ Recall = TP/(TP+FN)$$
 $$Precision = TP/(TP+FP)$$
 $$F1 = 2 Recall* Precision /(Recall + Precision)$$
 $$PF = FP/(FP+TN )$$
-$$Acc = (TP+TN)/(TP+TN+FP+FN)$$\
\ No newline at end of file
+$$Acc = (TP+TN)/(TP+TN+FP+FN)$$\