sergio-verduzco/draculab

A simulator of rate units with delayed connections

draculab.py is a Python module containing classes to simulate networks of rate units with delayed connections. Draculab is an acronym of Delayed-Rate Adaptively-Connected Units Laboratory.

The original purpose behind this simulator was to create neural networks that autonomously learn to control a simulated physical plant. Because of this, rate units and their synapses are dynamical systems operating in continuous time. Moreover, since instantaneous connections are unrealistic, there is a minimum connection delay, and all connection delays in the network are multiples of it.

The simulator has a relatively simple architecture, so that experienced Python users can understand its source code and make any required adjustments.

See the INSTALL.md file for installation instructions.

A fast way to get started is to follow the tutorials in the tutorial folder. Users who want a more systematic introduction can begin by reading the accompanying paper, and then following the tutorials.

The paper should be in the tutorial folder, in the draculab_manuscript.pdf file. Its reference is: Verduzco-Flores SO, DeSchutter E (2019) "Draculab: A Python simulator for firing rate neural networks with delayed adaptive connections " Front. Neuroinform. 13, https://doi.org/10.3389/fninf.2019.00018

The lessons of the tutorial are given as Jupyter notebooks with filenames tutorialX.ipynb, where X stands for the tutorial number. It is recommended to follow the lessons in order using the notebooks, but for those users who prefer not to use Jupyter, the Python source code and instructions are included in the tutorialX.py files.

The tests folder contains miscellaneous code to test various aspects of Draculab. Some basic unit tests are in unit_tests.py. Most unit tests have an associated Jupyter notebook named testX.ipynb, where X is some number. The testX notebooks run similar tests, but these may include plots and animations in order to facilitate debugging.

Draculab contains a research-grade implementation of a planar arm, based on the dynamics of the double pendulum. The planar_arm.ipynb file in tests contains code to visualize the arm and its muscles for a given set of insertion points. There is also code to run a simulation with the planar_arm class, and visualize the results using an animation where the arm and the muscle activations can be observed.

The double_pendulum_validation.ipynb notebook in tests shows how the double pendulum equations are derived using Sympy, and how those equations can be validated by comparing them with other derivations.