|
From: Alex M. <al...@ca...> - 2007-07-08 07:46:23
|
I'm hoping to use spasm for a quick ddf-like project whith gaussian and discrete distributions. I've just submitted a few changes which makes spasm-base compile with gcc-4.2. Also added a first stab at discrete 1-D pdf implementation. I have a couple of questions: 1. spasm-base defines a few low-level operators, these I understand (I think): a) pdf*pdf (combines two posteriors, the result is correct if they are independent) b) pdf*like (combines a prior and a likelihood, correct if cond. independent) c) pdf/pdf (removes some information from the posterior) d) conv(pdf,pdf) (prediction) but what are these? e) conv(pdf,pdf) (addition) f) conv(pdf,pdf) (subtraction) back to (d)... the fact that the two pdf's are of the same type (and size) suggests that the transition function x->x' does not depend on x: F(x,x')=f(x'), i.e. the same motion+noise is applied to the belief at all points of the state space. Is this true? My problem is that for discrete states (e.g. target type), these transitions are usually not constant and should be specified by a NxN matrix (where N is the size of the state vector). Is it possible to incorporate this? cheers, alex |