计算编辑距离（feed_dict 错误）

Question

我在 Tensorflow 中编写了一些代码来计算一个字符串和一组字符串之间的编辑距离。我无法找出错误。

import tensorflow as tf
sess = tf.Session()

# Create input data
test_string = ['foo']
ref_strings = ['food', 'bar']

def create_sparse_vec(word_list):
    num_words = len(word_list)
    indices = [[xi, 0, yi] for xi,x in enumerate(word_list) for yi,y in enumerate(x)]
    chars = list(''.join(word_list))
    return(tf.SparseTensor(indices, chars, [num_words,1,1]))


test_string_sparse = create_sparse_vec(test_string*len(ref_strings))
ref_string_sparse = create_sparse_vec(ref_strings)

sess.run(tf.edit_distance(test_string_sparse, ref_string_sparse, normalize=True))

此代码有效，当运行时，它产生输出：

array([[ 0.25],
       [ 1.  ]], dtype=float32)

但是当我尝试通过稀疏占位符输入稀疏张量来执行此操作时，出现错误。

test_input = tf.sparse_placeholder(dtype=tf.string)
ref_input = tf.sparse_placeholder(dtype=tf.string)

edit_distances = tf.edit_distance(test_input, ref_input, normalize=True)

feed_dict = {test_input: test_string_sparse,
             ref_input: ref_string_sparse}

sess.run(edit_distances, feed_dict=feed_dict)

这是错误回溯：

Traceback (most recent call last):

  File "<ipython-input-29-4e06de0b7af3>", line 1, in <module>
    sess.run(edit_distances, feed_dict=feed_dict)

  File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/client/session.py", line 372, in run
run_metadata_ptr)

  File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/client/session.py", line 597, in _run
    for subfeed, subfeed_val in _feed_fn(feed, feed_val):

  File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/client/session.py", line 558, in _feed_fn
    return feed_fn(feed, feed_val)

  File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/client/session.py", line 268, in <lambda>
    [feed.indices, feed.values, feed.shape], feed_val)),

TypeError: zip argument #2 must support iteration

知道这里发生了什么吗？

Answer 1

TL;DR: 对于 create_sparse_vec() 的 return 类型，使用 tf.SparseTensorValue instead of tf.SparseTensor.

这里的问题来自create_sparse_vec()的return类型，即tf.SparseTensor, and is not understood as a feed value in the call to sess.run()。

当您输入（密集）tf.Tensor 时，预期值类型是 NumPy 数组（或某些可以转换为数组的对象）。当您输入 tf.SparseTensor 时，期望值类型是 tf.SparseTensorValue，它与 tf.SparseTensor 类似，但它的 indices、values 和 [=23] =] 属性是 NumPy 数组（或某些可以转换为数组的对象，例如示例中的列表。

以下代码应该有效：

def create_sparse_vec(word_list):
    num_words = len(word_list)
    indices = [[xi, 0, yi] for xi,x in enumerate(word_list) for yi,y in enumerate(x)]
    chars = list(''.join(word_list))
    return tf.SparseTensorValue(indices, chars, [num_words,1,1])

计算编辑距离（feed_dict 错误）

Computing Edit Distance (feed_dict error)

sparse-matrix

levenshtein-distance

tensorflow