python - "Cannot convert a ndarray into a Tensor or Operation."尝试从 tensorflow 中的 session.run 获取值时出错

Question

我在 tensorflow 中创建了一个孪生网络。我正在使用以下代码计算两个张量之间的距离:

distance = tf.sqrt(tf.reduce_sum(tf.square(tf.subtract(question1_predictions, question2_predictions)), reduction_indices=1))

我能够毫无错误地训练模型。在推理部分，我正在检索 distance张量如下:

test_state, distance = sess.run([question1_final_state, distance], feed_dict=feed)

Tensorflow 抛出错误:

Fetch argument array([....], dtype=float32) has invalid type , must be a string or Tensor. (Can not convert a ndarray into a Tensor or Operation.)

当我打印 distance张量，在 session.run 之前和之后在训练部分，它显示为<class 'tensorflow.python.framework.ops.Tensor'> .所以张量的替换distance用 numpy distance正在发生在 session.run推理部分。按照推理部分的代码:

with graph.as_default():
    saver = tf.train.Saver()

with tf.Session(graph=graph) as sess:
    sess.run(tf.global_variables_initializer(), feed_dict={embedding_placeholder: embedding_matrix})
    saver.restore(sess, tf.train.latest_checkpoint('checkpoints'))

    test_state = sess.run(initial_state)

    for ii, (x1, x2, batch_test_ids) in enumerate(get_test_batches(test_question1_features, test_question2_features, test_ids, batch_size), 1):
        feed = {question1_inputs: x1,
                question2_inputs: x2,
                keep_prob: 1,
                initial_state: test_state
               }
        test_state, distance = sess.run([question1_final_state, distance], feed_dict=feed)

Answer 1

看起来你用 numpy 数组 distance = sess.run(distance) 覆盖了 Tensor distance = tf.sqrt(...)。

你的循环是罪魁祸首。将 t_state, distance = sess.run([question1_final_state, distance] 更改为 t_state, distance_other = sess.run([question1_final_state, distance]