加权交叉熵损失函数

TextCNN的loss函数

最近看到一篇论文通过设置带权重的交叉熵损失函数解决数据集中类不平衡问题，没有找到实现的代码，就参考了几篇代码写了一个。
论文链接: Neural Network-based Detection of Self-Admitted Technical Debt: From Performance to Explainability.

基于tensorflow2+python3.7的代码实现

import tensorflow as tf
from keras import backend as K
def get_weight(weights):
    def mycrossentropy(y_true, y_pred):
        pt_1 = tf.where(tf.equal(y_true, 1), y_pred, tf.zeros_like(y_pred))
        pt_0 = tf.where(tf.equal(y_true, 0), y_pred, tf.ones_like(y_pred))
        loss = (1-weights)*K.binary_crossentropy(y_true, y_pred)*pt_1+weights*K.binary_crossentropy(y_true, y_pred)*pt_0
        return loss
    return mycrossentropy
    
model.compile('adam', loss=get_weight(weights), metrics=['accuracy'])

参考代码来源：
Focal Loss