HTM prediction model with good accuracy but low activation levels

Engineering NuPIC
,
2

The active columns are defined by the Spatial Pooler, not the Temporal Memory. The TM activates cells within active columns. To get the active columns from the follow the instructions [here(NuPIC Usage FAQ), specifically:

# Extract the spatial pooler
spRegion = model._getSPRegion()
sp = spRegion.getAlgorithmInstance()

Use this API for the SP.

# Extract the temporal memory
tmRegion = model._getTPRegion()
tm = tmRegion.getSelf()._tfdr

Here is the API for the Backtracking TM.

# Get the active cells
tm.infActiveState["t"]

These active cells are in a 1d array. They need to be transformed into SP columns.

1 Like