CODEWITHSUNDEEP
pythonloopstensorflowslicetensor
LY-
LY-

Reputation: 23

Slice tensor using tf.while_loop

I'm trying to slice the tensor into small ones as long as there's still some columns using tf.while_loop.
Note : I'm using this way because I cannot loop over a value in a placeholder at the graph construction time ( without session ) considered as a tensor and not integer.

[ 5 7 8 ]      
[ 7 4 1 ]  =>  
[5 7 ]     [ 7 8 ]  
[7 4 ]     [ 4 1 ]

This is my code: 

i = tf.constant(0)

result = tf.subtract(tf.shape(f)[1],1)

c = lambda result : tf.greater(result, 0)

b = lambda i: [i+1, tf.slice(x, [0,i],[2, 3])] 

o= tf.while_loop(c, b,[i])

with tf.Session() as sess: 

   print (sess.run(o))

However, I get this error : 

 ValueError: The two structures don't have the same nested structure.

 First structure: type=list str=[<tf.Tensor 'while_2/Identity:0' shape=()                     dtype=int32>]
 Second structure: type=list str=[<tf.Tensor 'while_2/add:0' shape=()     dtype=int32>, <tf.Tensor 'while_2/Slice:0' shape=(2, 3) dtype=int32>]

I would like to return the sub tensor everytime

Upvotes: 1

Views: 743

Answers (1)

benjaminplanche
benjaminplanche

Reputation: 15119

There are several problems with your code:

  • You are not passing any structure/tensor to receive the values of your tf.slice(...). Your lambda b should have a signature such as lambda i, res : i+1, ...

  • Tensors edited through a tf.while_loop should have a fixed shape. If you want to build a loop to collect slices, then you should first initialize the tensor res with the appropriate shape to contain all the slice values, e.g. res = tf.zeros([result, 2, 2]).

Note: Regarding your particular application (collecting all pairs of neighbor columns), this could be done without a tf.while_loop:

import tensorflow as tf

x = tf.convert_to_tensor([[ 5, 7, 8, 9 ],
                          [ 7, 4, 1, 0 ]])
num_rows, num_cols = tf.shape(x)[0], tf.shape(x)[1]

# Building tuples of neighbor column indices:
n = 2 # or 5 cf. comment
idx_neighbor_cols = [tf.range(i, num_cols - n + i) for i in range(n)]
idx_neighbor_cols = tf.stack(idx_neighbor_cols, axis=-1)

# Finally gathering the column pairs accordingly:
res = tf.transpose(tf.gather(x, idx_neighbor_cols, axis=1), [1, 0, 2])

with tf.Session() as sess:
    print(sess.run(res))
    # [[[5 7]
    #   [7 4]]
    #  [[7 8]
    #   [4 1]]
    #  [[8 9]
    #   [1 0]]]

Upvotes: 1

Related Questions