如何在张量流中枚举张量？

Question

In Google's translate model file--seq2seq.py, they use 'enumerate' to enumerate a tensor in line 1195 as in the pictiure show, why when i try to enumerate a tensor but it shows 'TypeError: 'Tensor' object is not iterable.' and i can't found any clue about how to enumerate a tensor in tensorflow's tutorial,can anyone help me? Thanks! BY the way: my tensor is rank 4 enter image description here

Answer 1

You can slice and index tensors. The only difference is that you need to run a session to get the values. Here is a example of iterating

foo = tf.constant([1,2,3,4,5,6,7,8,9,0])
sess = tf.Session();
for i in range(10):
    print(sess.run(foo[i]))
sess.close()

Alternatively you can use a interactive session and directly call print(foo[i].eval())

or enable eager execution and it becomes as easy as iterating a Python list.

tf.enable_eager_execution()
foo = tf.constant([1,2,3])
for i in foo:
    print(i) #this prints the tensor including value

This prints

tf.Tensor(1, shape=(), dtype=int32)

tf.Tensor(2, shape=(), dtype=int32)

tf.Tensor(3, shape=(), dtype=int32)

In all above you will get a tensor not the exact value

This method does works for placeholders, constants and Variables.

Only thing : For eager mode in Variables, you need to use tf.contrib.eager.Variable , instead of tf.Variable