Re: Google's Gson JSON

[Jeff Eastman]

Sounds like we have a couple of additional alternatives to try. I'm going to continue with Gson in the DP MR stuff for the short term, since it is already working, and maybe try all three with Vector and Matrix as they stand now without annotations.

Jeff


Sean Owen wrote:

From a recent posting it sounds like annotations are bubbling up in priority.

If we're going this way -- and I strongly support it -- I'd suggest we
look a step beyond JSON. It is a more compact and standard string
encoding of complex data types, indeed. But it has the secondary goal
of being parseable as Javascript, and a string representation is not
the most efficient encoding.

This strikes me as exactly what Protocol Buffers (or Thrift from FB
perhaps) is for. It is certainly exactly what is used inside Google
for moving data around among MapReduces. It also has Java bindings.

On Mon, Mar 9, 2009 at 6:45 PM, Jeff Eastman <j...@windwardsolutions.com> wrote:

A few months back, in the context of vector annotations, we had a discussion
of a more standard means to serialize our object state. The Dirichlet
Process implementation has a rather complicated DirichletState object which
must be serialized and so I have worked out a way to do this using Gson.
Though I had to use the 1.3 beta 2 release to get past a problem in the
1.2.3 release, the package seems to be up to the task of serializing
complicated, generic, classes.

In the post 0.1 timeframe, I will look into using native Gson to replace the
current Vector asFormatString as a step towards vector annotations. It's
Apache licensed.

Does anybody else have experience with or comments about this package?

Jeff

PGP.sig
Description: PGP signature