The accompanying HTM reading list records the research background; accessing legacy Numenta content explains why some original installation references disappeared.
Early NuPIC binaries were linked against Python 2.5.4. During a project on a 64-bit openSUSE system, the installed Python was 2.6.2 and downgrading the distribution’s interpreter was not a sensible option. I tested a compatibility workaround by exposing the available Python libraries under the names expected by NuPIC.
After installing NuPIC according to Numenta’s then-current Linux instructions, I created the following symbolic links.
For the 64-bit libraries:
cd /usr/lib64/
sudo ln -s libpython2.6.so libpython2.5.so.1.0
sudo ln -s libpython2.5.so.1.0 libpython2.5.so.1
sudo ln -s libpython2.5.so.1 libpython2.5.so For the standard library path:
cd /usr/lib/
sudo ln -s libpython2.6.so.1.0 libpython2.5.so.1.0
sudo ln -s libpython2.5.so.1.0 libpython2.5.so.1
sudo ln -s libpython2.5.so.1 libpython2.5.so I also linked the interpreter directory and NuPIC’s own library directory:
cd /usr/local/lib/
sudo ln -s python2.6 python2.5
cd "$HOME/nta/lib"
ln -s python2.5 python2.6 The minimal import test was:
import nupic.network I tested the setup on openSUSE 11.2 and 11.3, on both 32-bit and 64-bit installations, and ran several NuPIC examples including the Vision Toolkit without an error attributable to the Python version.
This was a pragmatic compatibility hack, not a general claim that two interpreter ABIs are interchangeable. A symlink cannot repair a real ABI mismatch. On a modern system the sound approach would be an isolated, reproducible environment containing the exact historical runtime—or, preferably, a maintained version of the software.
Published at Internet Archive ↗