A Nengo model designed to count, learn and retrieve alphabet-arithmetic problems such as A+2=C and to respond YES or NO depending on the validity of the input problem.

The model was designed using Nengo (version 2.8.0).
Nengo requires python3 (version 3.6.8).

To improve computing time, the OpenCL version of nengo was used (nengo-ocl version 1.4.0).
nengo-ocl requires pyopencl (version 2018.1.1). The model failed to run using newer versions of pyopencl.

NOTE: It is advised to use the nengo-ocl version included in this repository, as it contains a few additional files.

The following files contain different stages of development for the Nengo model. final_model.py contains the main model.

The final version, able to calculate, memorise and retrieve problems of the form A+2=C

The problem can be presented in ll. 123-124.
For retrieval of a previously learned problem, 'load_from' must be active (l. 246).
To create new memory: must be simulated using shell command 'python3 final_model.py', which creates a test.npz file.
To learn with empty memory, remove 'load_from' (l.246).

In order to change learning and retrieval in the model actions (ll.305-329):
Learning is switched on by setting 'do_learn=GO'. Default is no learning (do_learn=CLEAR).
retrieval is switched on by setting 'do_dm=GO'. Default is no retrieval (do_dm=CLEAR).
Memory module receives input via 'in_layer' and 'correct' networks (see l. 324).

Intermediate version of the model. Is able to count and retrieve from a predefined memory, but is unable to create new memory. Yields better output than the final model.

Memory content can be predefined in ll. 112-114.
Input problem can be changed in ll. 117-1118.

Memory module to memorise and retrieve problems of the form A+2=C.

For recall:
Switch off ll. 127, 129, 139 and 142-145.
Make sure 'load_from' is active (l.133).
Input should have the form 'ITEM1*B + ITEM2*TWO' (ll. 71-91).
Recall is run in Nengo GUI.

For learning:
Switch on all learning parameters and connections (see recall).
Optional: switch off 'load_from' (l. 133) to use an empty memory.
Input should have the form 'ITEM1*B + ITEM2*TWO + RESULT*D' (ll. 71-91).
To store memory, simulate model using shell command 'python3 bcm_zbrodoff.py'. Duration of simulation can be set in l. 156. Must be at least 1 second long.

The count model before combining the different parts. Works for a limited range pf problems.
The problem can be defined in ll.58-59.

Model is only able to retrieve. Has all problems containing A-F and 2-4 stored in memory.
Problem can be defined in l. 100.

The count model as given to me as basis for the project.
The problem can be defined in l. 36.

Memory module as given to me as basis for this project. Not designed for alphabet-arithmetic problem.
Input over time can be set in ll. 24-28. Can memorise word-pairs containing the words 'CAT', 'DOG', 'CHAIR', and 'TABLE'.