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

First, to explain what that documentation means:

1. For dense data usually you are doing:

   Serialized Example (from reader) -> parse_single_example -> batch queue -> use it.

2. For sparse data you currently need to do:

   Serialized Example (from reader) -> batch queue -> parse_example -> use it.

An example of this would be:

reader  = tf.TFRecordReader()
_, serialized_example = reader.read(filename_queue)
batch_serialized_examples = tf.shuffle_batch([serialized_example], batch_size)
feature_to_type = {
  'label': tf.FixedLenFeature([1], dtype=tf.int64),
  'sparse_feature': tf.VarLenFeature(dtype=tf.int64)
}
features = tf.parse_example(batch_serialized_examples, feature_to_type)

Note, shuffle_batch takes a series of strings and returns batch of strings. label should be fixed len of rank == 1 from your example.