Open oxtopus opened 8 years ago
@oxtopus I trust you will close this issue when you are satisfied with an answer and open up any new actions that should be taken in response to the answer as new issues.
@oxtopus What is that, your theme song? 👏
Signal-boosting to @numenta/nupic-committers in case anyone has any opinions. If not, I'll roll some dice and make them consistent in a followup PR.
@rhyolight here's the mixtape version: https://www.youtube.com/watch?v=4YKlDpX3swI
The default parameters for the c++ SpatialPooler (https://github.com/numenta/nupic.core/blob/master/src/nupic/algorithms/SpatialPooler.hpp#L232-L249) differ from the swig interface (https://github.com/numenta/nupic.core/blob/master/src/nupic/bindings/algorithms.i#L1020-L1037) file that provides bindings for python. Furthermore, both differ from the pure-python implementation in nupic (https://github.com/numenta/nupic/blob/master/src/nupic/research/spatial_pooler.py#L111-L129).
i.e. if you instantiate SP, you get different default parameters depending on which implementation you use.
Particularly,
globalInhibition
,seed
,synPermActiveDec
differ between implementations, and some names differ, e.g.minPctOverlapDutyCycle
vsminPctOverlapDutyCycles
.Which is correct?