用于在keras中调用的自定义宏 [英] Custom macro for recall in keras

问题描述

我正在尝试为recall = (recall of class1 + recall of class2)/2创建一个自定义宏.我想出了以下代码，但不确定如何计算0类的正值.

I am trying to create a custom macro for recall = (recall of class1 + recall of class2)/2. I came up with the following code but I am not sure how to calculate the true positive of class 0.

def unweightedRecall():
    def recall(y_true, y_pred):
        # recall of class 1
        true_positives1 = K.sum(K.round(K.clip(y_pred * y_true, 0, 1)))
        possible_positives1 = K.sum(K.round(K.clip(y_true, 0, 1)))
        recall1 = true_positives1 / (possible_positives1 + K.epsilon())

        # --- get true positive of class 0 in true_positives0 here ---
        # Also, is there a cleaner way to get possible_positives0
        possible_positives0 = K.int_shape(y_true)[0] - possible_positives1
        recall0 = true_positives0 / (possible_positives0 + K.epsilon())
        return (recall0 + recall1)/2
    return recall

似乎我将不得不使用Keras.backend.equal(x, y)，但是如何创建具有形状K.int_shape(y_true)[0]和所有值(例如x)的张量?

It seems I will have to use Keras.backend.equal(x, y), but how do i create a tensor with shape K.int_shape(y_true)[0] and all values, say x?

编辑1

基于Marcin的评论，我想基于keras中的回调创建自定义指标.在浏览Keras中的问题时，我遇到了以下用于f1指标的代码:

Based on Marcin's comments, I wanted to create a custom metric based on callback in keras. While browsing issues in Keras, I came across the following code for f1 metric:

class Metrics(keras.callbacks.Callback):
    def on_epoch_end(self, batch, logs={}):
        predict = np.asarray(self.model.predict(self.validation_data[0]))
        targ = self.validation_data[1]
        self.f1s=f1(targ, predict)
        return
metrics = Metrics()
model.fit(X_train, y_train, epochs=epochs, batch_size=batch_size, validation_data=[X_test,y_test], 
       verbose=1, callbacks=[metrics])

但是回调如何返回准确性?我想实现unweighted recall = (recall class1 + recall class2)/2.我可以想到以下代码，但希望能有帮助完成它

But how is the callback returning the accuracy? I wanted to implement unweighted recall = (recall class1 + recall class2)/2. I can think of the following code but would appreciate any help to complete it

from sklearn.metrics import recall_score
class Metrics(keras.callbacks.Callback):
    def on_epoch_end(self, batch, logs={}):
        predict = np.asarray(self.model.predict(self.validation_data[0]))
        targ = self.validation_data[1]
        # --- what to store the result in?? ---
        self.XXXX=recall_score(targ, predict, average='macro')
        # we really dont need to return anything ??
        return
metrics = Metrics()
model.fit(X_train, y_train, epochs=epochs, batch_size=batch_size, validation_data=[X_test,y_test], 
       verbose=1, callbacks=[metrics])

---

模型:

---

Edit 2: model:

def createModelHelper(numNeurons=40, optimizer='adam'):
    inputLayer = Input(shape=(data.shape[1],))
    denseLayer1 = Dense(numNeurons)(inputLayer)
    outputLayer = Dense(1, activation='sigmoid')(denseLayer1)
    model = Model(input=inputLayer, output=outputLayer)
    model.compile(loss=unweightedRecall, optimizer=optimizer)
    return model

推荐答案

keras版本(存在问题).

keras version (with the mean problem).

您的两个类实际上仅是一维输出(0或1)吗?

Are your two classes actually only one dimension output (0 or 1)?

如果是这样:

def recall(y_true, y_pred):
    # recall of class 1

    #do not use "round" here if you're going to use this as a loss function
    true_positives = K.sum(K.round(y_pred) * y_true)
    possible_positives = K.sum(y_true)
    return true_positives / (possible_positives + K.epsilon())


def unweightedRecall(y_true, y_pred):
    return (recall(y_true,y_pred) + recall(1-y_true,1-y_pred))/2.

---

现在，如果您的两个类实际上是2元素输出:

---

Now, if your two classes are actually a 2-element output:

def unweightedRecall(y_true, y_pred):
    return (recall(y_true[:,0],y_pred[:,0]) + recall(y_true[:,1],y_pred[:,1]))/2.

---

回调版本:

对于回调，可以使用LambdaCallback，然后手动打印或存储结果:

For the callback, you can use a LambdaCallback, and you manually print or store the results:

myCallBack = LambdaCallback(on_epoch_end=unweightedRecall)
stored_metrics = []

def unweightedRecall(epoch,logs):
    predict = model.predict(self.validation_data[0])
    targ = self.validation_data[1]

    result = (recall(targ,predict) + recall(1-targ,1-predict))/2. 
    print("recall for epoch " + str(epoch) + ": " + str(result))
    stored_metrics.append(result)

其中recall是使用np而不是K的函数.还有epsilon = np.finfo(float).eps或epsilon = np.finfo(np.float32).eps)

Where recall is a function using np instead of K. And epsilon = np.finfo(float).eps or epsilon = np.finfo(np.float32).eps)

这篇关于用于在keras中调用的自定义宏的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！