2.3 Performance Metrics
2.3.1 Performance metrics in a regression taskmean square error
2.3. Performance metrics in the 2 classification tasksAccuracy acc = (TP+TN)/(TP+FN+FP+TN) error rate E =(FN+FP)/(TP+FN+FP+TN)
Accuracy P = tp/(TP+FP) The proportion of the melon picked out by the good melon is the recall rate R = tp/(TP+FN) = tp/m+ The ratio of good melons to all the good melons picked out.True sample Rate TPR = tp/(TP+FN) =tp/m+ The proportion of all positive cases selectedFalse positive case Rate FPR = fp/(TN+FP) =fp/m- The proportion of all counter -Examples selected
False counter Example Rate FNR = fn/(TP+FN) =fn/m+ The proportion of all positive cases that have not been singled outFNR+TPR = 1
P and R "mutually exclusive", a 1 1. Make up PR graph and PR curve. TPR andFPR "phase suction" at the same time for 0 at the same time as 1. The ROC graph and the ROC curve are formed.
F1 are the harmonic averages of P and r:1/f1 = (1/p + 1/r) =1/(2*P) + 1/(2*r)FB is a weighted harmonic average of P and r:1/FB = 1/(1+b^2) * (1/p +b^2/R) = 1/((1+b^2) *p) +b^2/((1+b^2) *R) =1/((1+b^2) *p) +1/((1+1/b^2) *R)
3.2 Linear regression
Null
"Machine learning" (Zhou Zhihua) notes
