adriaant / calmpy

Python implementation of a CALM neural network

CALMPY

A Python implementation of the CALM neural network using numpy. For an overview of CALM please see https://www.researchgate.net/publication/288303968_State_of_CALM_draft

Network topologies are defined using JointJS via a Django website, while simulations are done using command-line Django shell or scripts.

Quickstart

Editing networks

1. Create a virtual environment and install packages listed under requirements.
2. Activate the virtual env and run a standard Django migration.
3. Edit DATA_DIR so that it points to an existing location at your drive.
4. Optionally load a sample network under tests/fixtures.
5. Run python manage.py runserver and navigate to http://127.0.0.1:8000/ which shows a stock interface. The only functional part is the section named "Available networks".
6. When in the network editor interface, use the control menu to define modules.

Running simulations

1. Assumming you're using the sample network named 'simple', first ensure input patterns are defined. You can copy the 'simple' directory in tests/data to the DATA_DIR destination.
2. Then, in the terminal, start a shell with python manage.py shell and run, e.g:

    from simulator import load_network
    network = load_network('simple')
    network.train(50, 50)
    network.display()
    network.performance_check()

3. A 2D plot with categorization performance can be created with:

    from simulator.tools import *
    c = ConvergenceMap(network)
    c.display('in', 0, 1, 'out', 100)

Note: You need to install gnuplot for this. On OS X: brew install --with-aquaterm --with-x11 --without-emacs --with-cairo --with-qt --with-pdflib-lite gnuplot

(more documentation to follow)