Tensorflow计算正确率、精确率、召回率

二分类模型的评价指标

https://www.cnblogs.com/xiaoniu-666/p/10511694.html

参考tf的方法

1predictions = tf.argmax(predict, 1)
2    actuals = tf.argmax(real, 1)
3
4ones_like_actuals = tf.ones_like(actuals)
5    zeros_like_actuals = tf.zeros_like(actuals)
6    ones_like_predictions = tf.ones_like(predictions)
7    zeros_like_predictions = tf.zeros_like(predictions)
8
9Lable:      1   1   0   0
10        predi:      1   0   0   1
11                    Tp  Fp  Tn  Fn
12    tp: = and       1
13    tn = ont(or)            1
14
15    lab-pred:       0   1   0   -1
16
17    lab-pred>=0.6:  0   1   0   0
18    fp = and(lable, lab-pred):
19                    0   1   0   1
20
21    lab-pred<=-1.0: 0   0   0   1
22    not-lable:      0   0   1   1
23    fn = and(not-lable, lab-pred<-1.0)
24
25可能用到的方法：
26
27tf.less_equal
28tf.less
29tf.greater_equal
30tf.greatercount_nonzero

参考：

https://blog.csdn.net/sinat\_35821976/article/details/81334181

https://tensorflow.google.cn/api\_docs/python/tf/math/count\_nonzero