Continuous-Time, Dynamic Implementation of Numenta's Hirarchical Temporal Memory Networks

Experiments

Spatial Pooler

Temporal Memory

Sequences

The sequence experiment for the temporal memory implements a small LIF based temporal memory network with 128 columns and 8 cells each. However, no actual learning rules are implemented. Therefore, the connectivity of the model must be set externally.

At first, a NuPIC temporal memory instance is trained with a predifined stimulus. It's connectivity can be dumped using the extract.py script:

$ python2 extract.py

You will find the files stimulus.npy, labels.npy, and connectivity.npy in your directory. You might want to get familiar with the script's command line arguments by simply appending --help to the command string.

The actual simulation is initiated by running

$ python2 sequences.py

Again, check the command line help for options.