TensorFlow SparseTensor with dynamically set dense_shape

Question

I have previously asked this question Create boolean mask on TensorFlow about how to get a tensor only with certain indices set to 1, and the rest of them to 0.

I thought the answer given by @MZHm would entirely solve my problem. Although, the argument dense_shape of tf.SparseTensor only accepts lists, and I want to pass a shape which is inferred from the graph (from the shape of another tensor which has variable shape). So in my case in specific, I want to do something like this:

# The tensor from which the shape of the sparse tensor is to be inferred
reference_t = tf.zeros([32, 50, 11])

# The indices that will be 1
indices = [[0, 0],
           [3, 0],
           [5, 0],
           [6, 0]]

# Just setting all the values for the sparse tensor to be 1
values = tf.ones([reference_t.shape[-1]])

# The 2d shape I want the sparse tensor to have
sparse_2d_shape = [reference_t.shape[-2],
                   reference_t.shape[-1]]

st = tf.SparseTensor(indices, values, sparse_2d_shape)

From this I get the error:

TypeError: Expected int64, got Dimension(50) of type 'Dimension' instead.

How to I dynamically set the shape of a sparse tensor? Is there a better alternative to achieve what I'm aiming to do?

Answer 1

Here is what you can do to have a dynamic shape:

import tensorflow as tf 
import numpy as np

indices = tf.constant([[0, 0],[1, 1]], dtype=tf.int64)
values = tf.constant([1, 1])
dynamic_input = tf.placeholder(tf.float32, shape=[None, None])
s = tf.shape(dynamic_input, out_type=tf.int64)

st = tf.SparseTensor(indices, values, s)
st_ordered = tf.sparse_reorder(st)
result = tf.sparse_tensor_to_dense(st_ordered)

sess = tf.Session()

An input with (dynamic) shape [5, 3]:

sess.run(result, feed_dict={dynamic_input: np.zeros([5, 3])})

Will output:

array([[1, 0, 0],
       [0, 1, 0],
       [0, 0, 0],
       [0, 0, 0],
       [0, 0, 0]], dtype=int32)

An input with (dynamic) shape [3, 3]:

sess.run(result, feed_dict={dynamic_input: np.zeros([3, 3])})

Will output:

array([[1, 0, 0],
       [0, 1, 0],
       [0, 0, 0]], dtype=int32)

So there you go... dynamic sparse shape.