Skip to content

Recurrent network of leaky integrate-and-fire neurons

Notifications You must be signed in to change notification settings

RomanKoshkin/LIF

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Parallelization

Compiled with #pragma omp parallel for in front of the main loop of ONE time step YOU CAN'T PARALLELIZE CODE OVER TIME

Compile with

g++ -std=gnu++11 -Ofast -ftree-vectorize -march=native -mavx -fopenmp LIF-2.cpp -o lif - On the Mac, OpenMP is not installed by default. Install it separately.

Numba is awesome

don't

- use huge matrices (3d especially) to store data
- try to 'help' Numba to parallelize the code by creating loops
- do the same in c++ and #pragma

numba is even slightly better than super optimized c++ code

- uses the cores efficiently

however, if you complile with std=gnu++17 (not 11) you'll beat numba

  • on a 32 core Intel machine -- 3000 neurons - 1 minute per 100 ms of simulation with dt=0.01
  • on a 128 core AMD ROC node -- 3000 neurons - 1 minute per 1000 ms of simulation with dt=0.01

Most recent code

LIF-4.cpp

- latest with STP
- embedded assemblies

LIF_numba_parallel-2.py

- same as LIF-4.cpp
- almost as fast
- easier to debug

Plots.ipynb

- let's you plot the rasters and weights from LIF-4.cpp

Illustration of the LIF network DONT CHANGE.ipynb

- essentially a tutorial, where you can plot AMPA, NMDA, GABA and other dynamics of a network of LIF neurons. The code is essentially the same as LIF_numba_parallel-2.py