如何在tensorflow线性链式CRF中正确传递transition_params的初值

import tensorflow as tf
import tensorflow_addons as tfa

def create_model():
    # 定义输入和标签占位符
    inputs = tf.placeholder(dtype=tf.float32, shape=[None, num_steps, num_features], name='inputs')
    labels = tf.placeholder(dtype=tf.int32, shape=[None, num_steps], name='labels')

    # 定义CRF层
    crf = tfa.layers.CRF(num_labels)

    # 计算CRF层的输出和损失函数
    logits = crf(inputs)
    loss = crf.loss(labels, logits)

    # 定义优化器和训练操作
    optimizer = tf.train.AdamOptimizer(learning_rate=0.001)
    train_op = optimizer.minimize(loss)

    # 返回模型的输入、输出和训练操作
    return inputs, labels, logits, loss, train_op

# 创建模型
inputs, labels, logits, loss, train_op = create_model()

# 初始化变量
init_op = tf.global_variables_initializer()
sess.run(init_op)

# 获取CRF层的transition_params
transition_params = crf.get_transition_params()

# 打印transition_params的初值
print("Initial transition_params:", sess.run(transition_params))

# 训练模型
for i in range(num_epochs):
    # 获取batch数据
    batch_inputs, batch_labels = get_batch_data()

    # 执行训练操作，并传递transition_params
    _, batch_loss, batch_transition_params = sess.run([train_op, loss, transition_params], 
                                                     feed_dict={inputs: batch_inputs, labels: batch_labels, 
                                                                crf.transition_params: batch_transition_params})

    # 打印每个epoch的损失和transition_params
    print("Epoch:", i, "Loss:", batch_loss)
    print("Transition_params:", batch_transition_params)