Input would be a discrete field, with arguments on number of hash buckets, and output is the result of `tf.string_to_hash_bucket_fast()`.