In this paper we consider the problem of storing sequences of symbols in
a compressed format, while supporting random access to the symbols without
decompression. Although this is a well-studied problem when the data is
textual, the kind of sequences we look at are not textual, and we argue
that traditional compression methods used in the text algorithms community
(such as compressors targeting $k$-th order empirical entropy) do not
perform as well on these sequential data, and simpler methods such
as Huffman-coding the deltas between sequence elements give better
compression performance. We discuss data structures that allow
random access to sequence elements that target such measures.