On 13 September 2011 13:13, Ted Dunning <[email protected]> wrote:
> The cluster, classification and decompositional jobs all like the same kind
> of input.  These can be viewed as matrices or sequences of vectors; it comes
> to much the same sort of thing.  The gotcha is that the user often has
> tokens in fielded documents (ratings, documents, purchase history).  Other
> than that, it should be pretty easy.  Even the output of most of these
> programs can be matrices/vector sequences.

Reminds me of visual compositing tools (eg Blender,
http://www.blender.org/development/release-logs/blender-242/blender-composite-nodes/
or Quartz Composer) where the "plumbing" that connects nodes with
pipes is just guided by input/output types, ... but that sort of basic
hint can't guarantee you've not hooked things up in a way that doesn't
really make sense or do what you hope. I'd be wary of putting too
heavy expectations on input/output typing here, it's pretty healthy
that the whole pipeline can be thought of as just communicating
vectors/matrices. Rather that, than try to guarantee against screwups
by strongly typing everything.

cheers,

Dan

Reply via email to