没有注册任何OpKernel支持这些属性的Op'HashTableV2'。 注册设备：[CPU，GPU]，注册内核：

Question

I am coding with tensorflow 1.5.0, python 3.5. I want to create a hashtable. Since I intend to assign values to it later, I create it in the init function like this.(the values and shape are randomly given) enter image description here

but then I encounter a problem like this enter image description here

Can anyone help me?

Answer 1

It seems that the implementation of HashTable in your version of TensorFlow does not provide kernels for every possible combination of key and value types. There are two things you can do:

• According to your error message, there is a kernel implementation for 64-bit integer keys and 32-bit float values. So one possible fix is to simply change the data type of keys to tf.int64 :

   keys = tf.constant([1, 2, 3]), dtype=tf.int64) 

• Another possibility is to update TensorFlow to a version where this combination of key and value is implemented. It seems this was added in version v1.11.0-rc0 ( see commmit ), so upgrading to that or a later version (in general it is more recommendable to upgrade to a stable version instead of a release candidate) should also fix the problem.

Answer 2

The answers by @jdehesa is really great. It works for me!!! my tf version is 1.4, python=3.6

Here is my code which works:

import tensorflow as tf
from tensorflow.contrib.lookup import *

k = tf.range(1, 3, dtype=tf.int64)
v = tf.range(5, 7, dtype=tf.int64)
table = tf.contrib.lookup.HashTable(
    tf.contrib.lookup.KeyValueTensorInitializer(k, v, key_dtype=tf.int64, value_dtype=tf.int64), -1)

out = table.lookup(tf.constant([2,1], dtype=tf.int64))

with tf.Session() as sess:
    print(sess.run([k, v]))
    table.init.run()
    print(out.eval())