How to load sparse data with TensorFlow?

Question

There is a small snippet about loading sparse data but I have no idea how to use it.

SparseTensors don\'t play well with queues. If you use SparseTensors

Answer 1

Store indices and values in your TFRecords Examples, and parse with SparseFeature. For example, to store and load a sparse representation for:

[[0, 0, 0, 0, 0, 7],
 [0, 5, 0, 0, 0, 0],
 [0, 0, 0, 0, 9, 0],
 [0, 0, 0, 0, 0, 0]]

This creates a TFRecords Example:

my_example = tf.train.Example(features=tf.train.Features(feature={
    'index_0': tf.train.Feature(int64_list=tf.train.Int64List(value=[0, 1, 2])),
    'index_1': tf.train.Feature(int64_list=tf.train.Int64List(value=[5, 1, 4])),
    'values': tf.train.Feature(int64_list=tf.train.Int64List(value=[7, 5, 9]))
}))
my_example_str = my_example.SerializeToString()

And this parses it with SparseFeature:

my_example_features = {'sparse': tf.SparseFeature(index_key=['index_0', 'index_1'],
                                                  value_key='values',
                                                  dtype=tf.int64,
                                                  size=[4, 6])}
serialized = tf.placeholder(tf.string)
parsed = tf.parse_single_example(serialized, features=my_example_features)
session.run(parsed, feed_dict={serialized: my_example_str})

## {'sparse': SparseTensorValue(indices=array([[0, 5], [1, 1], [2, 4]]),
##                              values=array([7, 5, 9]),
##                              dense_shape=array([4, 6]))}

More exposition: Sparse Tensors and TFRecords