Thanks to this blog, and finally I get to know.
https://blog.csdn.net/hsqyc/article/details/81702437
1.TP TN FP FN concept
also read a lot about their interpretation of the concept before, but I felt did not understand thoroughly.
T N is represented or whether the sample points to be classified, P or N represents the sample is divided Why
TP (True Positives) to translate the meaning of our backwards is "is divided into positive samples and points to the" TN (True Negatives) means "to be divided into negative samples, but also points the way", FP (False Positives) means "to be divided into positive samples, but points wrong", FN (False Negatives) means "to be divided the sample is negative, but the points are wrong. "
For a classification task, it must have positive and negative samples, for example, to divide car class, then the car is positive samples, non-car are all negative samples.
As shown above, the left rectangle sample is positive, the right half rectangle is negative samples. A second sorter, in FIG drew a circle, a circle is considered to be within the classifier positive samples, the outer circle is negative samples. Then,
left semicircle classifier considered positive samples, while it is indeed a positive sample, it is "to be divided into positive samples and points the way" that TP.
Left rectangle deduct part is left semicircle classifier think it is a negative sample, but it itself is positive samples, that is, "is divided into negative samples, but points wrong" that FN.
Right semicircle classifier think it is a positive sample, but the sample itself is negative, then it is "is divided into positive samples, but points wrong" that FP.
The right half of the rectangular part is deducted right semicircle classifier think it is a negative sample, and the sample itself is indeed negative, then it is "is divided into negative samples, but also points to the" that TN.
2.Precision (precision) and Recall (recall) the concept of
classification considered positive category and is indeed part of the regular class accounted for all classifiers think is the proportion of positive class ", a measure of a classifier points out the positive class is indeed a positive class the probability two extremes is, if accuracy is 100 percent, on behalf of all classifiers points out the positive class are indeed positive class. If the precision is 0%, on behalf of the classifier points out the positive class is not a positive kind. just how well can not measure the accuracy of a classifier, such as the 50 positive samples and 50 negative samples, I classifier to 49 positive samples and 50 negative samples were divided into negative samples, leaving a positive sample points positive sample, so my accuracy is 100%, but a fool knows this classifier is garbage.
translated into Chinese is "classifier considered positive category and is indeed part of the regular class accounted for all is indeed the proportion of positive class", a measure of a classification of all ability can find out all the positive class. two extreme cases, if the recall rate was 100 percent, on behalf of all regular classes are divided into positive class classifier. If the recall rate is 0% Not on behalf of a class being divided into regular classes.