So I’m trying to get the prediction density out from NuPIC models.
The idea is to have a standard measure of model precision – along with the measure of surprise provided by the anomaly metrics.
After digging around the source, here’s my current approach.
Do you see anything off about this?
I’m currently getting prediction densities (“predCount” values) that mismatch with the anomaly scores.
For example anomaly score 0.75 may come with predCount 0.0 – which doesn’t make sense.
Thanks very much again