如何在 Keras 中的自定义损失函数中计算 True Positive

Question

I do classification task with Keras, I make simple custom loss function in Keras and it works

import keras.backend as K

def customLoss(yTrue,yPred):

    return K.abs(yTrue-yPred)

to make more complex loss function that i want, i need to calculate True Positive, True Negative , False Positive , False Negative

How to calculate them ?

i cant calculate them because i dont know the type of yTrue and yPred . Are they 2D array or list or anything else. if i know, maybe i can calculate TP,TN,FP,FN using for , like this:

TP=0
for x,y in zip(yTrue,yPred):
   if x == 1 and y > 0.5:
      TP=TP+1

Answer 1

According to the Keras Documentation the data types of yTrue/yPred are TensorFlow/Theano tensor depending on the backend you are using.

Therefore, you cannot use a for loop for the loss function, otherwise you will get an error.

But you can use logical and for this matter:

TN = np.logical_and(K.eval(y_true) == 0, K.eval(y_pred) == 0)
FP = np.logical_and(K.eval(y_true) == 0, K.eval(y_pred) == 1)

After that you can add them up:

TN = K.sum(K.variable(TN))
FP = K.sum(K.variable(FP))