[ml] Hacking tonight! 8PM, 83C

Michael C. Toren mct at toren.net
Wed Aug 26 20:39:53 UTC 2009


On Wed, Aug 26, 2009 at 12:23:25PM -0700, James Shearer wrote:
> Guys, sorry, it's not happening for me tonight.  This doubly sucks since
> this is likely my last wed. for awhile.

...and I'm in Boston this week.  Foo.

Okay, how about we try to work through some of the integration issues in
email?

Examining the Java code, the only place I see that calls into the
Quantizer (k-means) class is GestureModel.java, in the train() function.
It calls Quantizer.trainCenteroids(), gets the results with
Quantizer.getObservationSequence(), and feeds them to HMM.train().

Jim, your C port of train() takes a pointer to a list of StateSequence
structs.  StateSequence struct looks very similar to my observation
struct:

	wii_gestures/hmm/hmm.h:

		typedef struct _stateSequence {

			/* the length of this sequence */
			uint  length;

			/* the array of states in the sequence */
			uint* states;

		} StateSequence;
		typedef StateSequence* StateSequenceRef;

	wii_gestures/wiigee-c/quantizer.h:

		typedef struct observation {
		    int *sequence;
		    int sequence_len;
		} observation;

So, on the surface at least, it looks like "integrating" our two
components might be as simple as normalizing the above two data
structures?

-mct



More information about the ml mailing list