How to convert a tensor into a sparse integer encoding in Python

Question

I need to implement a function that converts a TF tensor of probability distributions to sparse integer encoding of categories. I don't think I'm going about it the right way. How do I get the expected output that I'm looking for?

My function:

def into_sparse(x):
    sparseData = tf.sparse.from_dense(x)
    return sparseData

Function call:

y_sparse = into_sparse(y)

Input (y):

tf.Tensor(
[[0.9933 0.     0.0067]
 [0.5065 0.1863 0.3072]
 [0.0751 0.9148 0.0102]
 [0.4307 0.0432 0.5261]], shape=(4, 3), dtype=float32)

Output:

<tensorflow.python.framework.sparse_tensor.SparseTensor at 0x7f3543aaa610>

Expected Output:

<tf.Tensor: shape=(4,), dtype=int64, numpy=array([0, 0, 1, 2])>

jkr · Accepted Answer · 2021-03-08 14:30:10Z

1

You can use argmax. That will give you the index that contains the maximum value in a dimension.

tf.math.argmax(y_sparse, axis=-1)

See https://www.tensorflow.org/api_docs/python/tf/math/argmax

This is different from the sparse arrays in tf.sparse. Those arrays have many zeros, and storing them in a sparse format rather than dense can save space.

edited Mar 8, 2021 at 14:30

answered Mar 8, 2021 at 3:38

jkr

19.6k5 gold badges49 silver badges78 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

How to convert a tensor into a sparse integer encoding in Python

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related