如何在Keras中将损失函数指定为二次加权kappa？

import keras.backend as K

def quadratic_weighted_kappa(y_true, y_pred):
    y_true = K.cast(K.argmax(y_true, axis=-1), dtype='int32')
    y_pred = K.cast(K.argmax(y_pred, axis=-1), dtype='int32')

    # Confusion matrix
    confusion = K.confusion_matrix(y_true, y_pred, num_classes=NUM_CLASSES)

    # Weight matrix
    weight_matrix = K.expand_dims(K.arange(0, NUM_CLASSES), axis=1) - K.expand_dims(K.arange(0, NUM_CLASSES), axis=0)
    weight_matrix = K.pow(weight_matrix, 2) / K.pow(NUM_CLASSES - 1, 2)

    numerator = K.sum(confusion * weight_matrix)
    denominator = K.sum(K.dot(K.sum(confusion, axis=0, keepdims=True), K.sum(confusion, axis=1, keepdims=True)) / K.sum(confusion))

    kappa = 1 - (numerator / denominator)
    return kappa

model.compile(loss=quadratic_weighted_kappa, optimizer='adam')

from keras.utils import to_categorical

# 将训练集和验证集的标签进行One-Hot编码
y_train = to_categorical(y_train, num_classes=NUM_CLASSES)
y_val = to_categorical(y_val, num_classes=NUM_CLASSES)

# 训练模型
model.fit(X_train, y_train, validation_data=(X_val, y_val), epochs=10, batch_size=32)