You can use any of the three implementations by specifying the "spatialImp" parameter in the "spParams" section when creating a model (or in description.py). The options are "oldpy", "py", and "cpp". The old Python version is still the default but we plan to switch to the C++ version in the next couple weeks, at which point we will completely remove the old Python version.
So if you have saved models created with the old Python version, you will not be able to deserialize them once the old Python code is deleted. Plan accordingly! On Fri, Sep 20, 2013 at 2:17 PM, Gil Shotan <[email protected]>wrote: > Hi all, > > For the past summer I've been working as an intern at Grok to create a > cleaner, more efficient implementation of the Spatial Pooler. I've finally > concluded my work and it has been merged to the master branch. One of the > purpose of my work was to help the community understand the code better, so > I simplified things, added a lot of comments, and adjusted names to make > the code conform to CLA white paper terminology. If you are interested in > the implementation details of the CLA please check out the new python > spatial pooler here: > > > https://github.com/numenta/nupic/blob/master/py/nupic/research/spatial_pooler.py > > I've also created an identical c++ implementation that should help us > move closer to the goal of having a stand alone c++ implementation of the > algorithms. It is also 20x faster than the current implementation. I > attached the slides of the presentation I gave here at grok about it. Feel > free to reach out if you have any questions, comments or feedback! > > Gil. > > _______________________________________________ > nupic mailing list > [email protected] > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org > >
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
