Python 如何使用tensoflow最小化由两个模型变量组成的损失

类模型\u DML（对象）：

train.py的关键代码：

        for i in range(MODEL_NUM):
           model = model_DML(layer=LAYER, n_users=user_num, n_items=item_num, emb_dim=EMB_DIM, lr=LR, 
                          lamda=LAMDA,optimization=OPTIMIZATION, 
                          pre_train_latent_factor=pre_train_feature, 
                          if_pretrain=IF_PRETRAIN,margin=Margin)
           models.append(model)
        train_batch_data = np.array(train_batch_data)
        kl_one = kl_div(models[0].cos_value,models[1].cos_value)
        kl_two = kl_div(models[1].cos_value,models[0].cos_value)
        loss_one = models[0].loss + kl_one
        loss_two = models[1].loss + kl_two
        opt_one = tf.train.AdamOptimizer(learning_rate=LR)
        opt_two = tf.train.AdamOptimizer(learning_rate=LR)
        updates_one = opt_one.minimize(loss_one)
        updates_two = opt_two.minimize(loss_two)
        sess.run(tf.global_variables_initializer())
        sess.run(tf.local_variables_initializer())

我想构造一个损失，就像某种知识提炼。 如何使损失1和损失2最小化？ kl_1和kl_2由两个模型的变量组成

        for i in range(MODEL_NUM):
           model = model_DML(layer=LAYER, n_users=user_num, n_items=item_num, emb_dim=EMB_DIM, lr=LR, 
                          lamda=LAMDA,optimization=OPTIMIZATION, 
                          pre_train_latent_factor=pre_train_feature, 
                          if_pretrain=IF_PRETRAIN,margin=Margin)
           models.append(model)
        train_batch_data = np.array(train_batch_data)
        kl_one = kl_div(models[0].cos_value,models[1].cos_value)
        kl_two = kl_div(models[1].cos_value,models[0].cos_value)
        loss_one = models[0].loss + kl_one
        loss_two = models[1].loss + kl_two
        opt_one = tf.train.AdamOptimizer(learning_rate=LR)
        opt_two = tf.train.AdamOptimizer(learning_rate=LR)
        updates_one = opt_one.minimize(loss_one)
        updates_two = opt_two.minimize(loss_two)
        sess.run(tf.global_variables_initializer())
        sess.run(tf.local_variables_initializer())