virtual evidence in a dynamic network

[con_fuse9]

We have a problem that we think will be best solved with a dynamic network. We are watching sensor data with a known, but varying uncertainty. We expect the uncertainty to change over time as well as the behavior of the target. When we descretize the data, given the uncertainty, the value typically follows a normal distribution over more than one value - hence the virtual evidence.

We would like to feed a dynamic network virtual evidence but can't find any easy way of doing it.

At this stage, we don't know how many slices (and would be dynamically adding slices as necessary). For convenience I don't want to 'unroll' the network and add groups of nodes to the end (although I suppose I could), nor if I decide to limit the network to fewer slices, do I want to have to chop off extra nodes.

I suspect that I could possibly characterize the problem differently, but this is a first effort.

I don't see this capability in SMILE, but I was wondering if it could be added? Could this be put on the road map?

Thanks

[shooltz[BayesFusion]]

I don't see this capability in SMILE, but I was wondering if it could be added?

It is actually in SMILE, as a slightly misleading overload of DSL_nodeValue::SetTemporalEvidence. The method you're looking for has the following signature:

Code: Select all

int SetTemporalEvidence(int slice, const std::vector<double> &evidence);

There's corresponding getter function as well.