We are moving towards a full C++ implementation. There was a thread where Gil described his efforts around the spatial pooler. There are some other things that would need to be ported to C++ like part of the temporal pooler, the encoders, and a CLA wrapper around the different pieces. Would be great if you were interested in contributing to that David, let me know and we can discuss.
With respect to understanding what is going on, it is possible to get the state out despite the hybrid Python/C++ implementation. We have a tool that we haven't made public yet called Cerebro. It provides step-by-step inspection of the CLA including the active and predicted cells. It also allows you to write a function for generating data to step through so that you can quickly see how the CLA learns and predicts for different artificial data sets. Cerebro is still tangled with our internal code but we understand it is an important tool for understanding the CLA so we would like to get it into the public repo as soon as possible. But no ETA or specific plans yet. On Mon, Jul 15, 2013 at 3:20 PM, David Ragazzi <[email protected]>wrote: > It could there's some plan to discard Python dependence. Because of > Python, several hacks/dependent tools are need for finally CLA run. > > I even am converting the entire Python code to C# and later to (native) > C++ in order to have a compiled stand-alone library which is cross platform > and so I could use it in my personal projects. It´s a plenty of work, but > for me is better. > > With CLA ported as stand-alone library anyone could develop independent > tools for different operating systems without need of virtualbox or others. > > I would love to know if Grook plans to remove the dependency on python > soon. I remember that someone here said something about, but I don't > remember exactly. If so, I would like much can help. > > Best, David > > > On 15 July 2013 18:26, Uwe Kirschenmann <uwe.kirschenmann@fit.** > fraunhofer.de <[email protected]>> wrote: > >> Hello, >> >> i am pretty new to the nupic mailing list but do have a few questions >> concerning the implementation of NuPIC and the documentation. It tooks some >> time to get the whole setup running on my virtual box and i let out a cry >> of joy when the tests endend sucessfully. i then set up eclipse and loaded >> the relevant python (OPF) files of the build and the experiment into my >> project folder. i debugged that to get an understanding and wrote down >> sequence diagrams (hotgym-experiment). from there i learned that some calls >> are made into the c++ libs via swig. Then i asked myself how to debug this >> code from within the calling python code - which i did not manage. what i >> can do, is call the c++ code from the sample tests you mentioned in the >> installation instructions. all in all that was pretty frustrating. >> >> my questions are: >> * do you have project-files that one can use (like in eclipse) for >> setting the structure up correctly? >> * is it possible to debug the c++ code from python that makes calls to >> swig-implementations? >> * what else is needed to get a better understanding of the whole setup >> (my guess is, that the participants of the hackathon do have some >> advantage), for example other directories in the numenta folder... >> >> Best regards, >> >> Uwe >> >> ______________________________**_________________ >> nupic mailing list >> [email protected] >> http://lists.numenta.org/**mailman/listinfo/nupic_lists.**numenta.org<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 > >
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
