mrg7 / hacktm Goto Github PK

HackTM C++ library

This library implements the Cortical Learning Algorithm as described
by Numenta's paper[1] in C++. It is meant to be a framework to
experiment with HTMs and the algorithm itself. 

This library works using bitmaps for input and output, and it is able
to associate these bitmaps to a generic, n-dimensional space.

No aggressive optimizations have been pursued, memory usage is not
optimal but it is fast enough and it is designed to be easily
pipelined by parallel threads.

To ease portability, it only uses STL library, with the exception of
LLVM BitVector class, included in the tree.


COMPILATION INSTRUCTIONS:

You will need cmake (http://www.cmake.org) to generate makefiles.
When you have installed cmake in your system, create a build
directory, cd into it and type:

 $ cmake /path/to/sources/

where "/path/to/sources/" is the place where the source have been
installed. If successful, you can type "make" and have the library and
tests compiled.


DIFFERENCES BETWEEN NUMENTA'S ALGORITHM AND HACKTM:

1. Boosting has not been implemented. There's no technical problem
behind implementing it, just lack of interest at this moment.
2. Sequence segments and prediction segments are treated
separately. This helps the sequential learner to better differentiate
between sequences with the same elements but in different order.


IMPORTANT NOTES ABOUT THE USE OF THIS SOFTWARE AND THE CLA ALGORITHM

This software is inteded for personal use and research only. Numenta
owns the Intellectual Property of the HTM algorithm. Please refer to
Numenta website for more information.

Also note that I am not affiliated at all with such company, and that
this library is the result of programming in my free time.


[1] http://numenta.com/htm-overview/education/HTM_CorticalLearningAlgorithms.pdf