Easy way of getting predictive cells from a network model?

Engineering NuPIC
1

Hi again,

Currently working on a network model that looks like this

Sensor -> SP -> TM -> next prediction from TM

However, I’m currently having major issue getting the output from the temporal memory. When I’m only working with the temporal memory object (not in a network structure) I can just call on the method getPredictiveCells() and get what I want. Any suggestions to do the same with a network model? I have tried “tm.getOutputData(“predictedActiveCells”)” but not really the same.

Cheers!

2

Have you tried tm.getSelf()? Inspect the object you get back and see if you can find the TM algorithm instance on it. Then you can call getPredictiveCells() on it.

3

Yeah, but I get into the problem of the region not having the attribute, i.e. AttributeError: ‘TMRegion’ object has no attribute ‘getPredictiveCells’. Tried to find the instance but by digging in the source code of tm_region.py and I found that (line 319-320)

Which Temporal implementation?

TemporalClass = _getTPClass(temporalImp)

you cannot access the TM instance. I guess a work around could be to change it to self.TemporalClass?

4

Try tmRegion.getAlgorithmInstance().getPredictiveCells()

1 Like
5

Worked like a charm! Thanks again Matt!