2023-08-02__speedtest_brian_standalone_AdEx_Nto1¶

%%time
from brian2 import *

CPU times: total: 875 ms
Wall time: 2.03 s

!mkdir cpp

A subdirectory or file cpp already exists.

set_device('cpp_standalone', directory='cpp/1')

%run lib/neuron.py

importing mpl … ✔
importing brian … ✔
importing pandas … ✔

μₓ = 4 * Hz
σ = sqrt(0.6)
μ = log(μₓ / Hz) - σ**2 / 2

1.0862943611198905

we = 14 * pS
wi = 4 * we
T = 10*second;

def Nto1_merged(N = 6500, Ne_simmed = 100, print_N=True):
    
    Ni_simmed = Ne_simmed

    Ne = N * 4//5
    Ni = N - Ne
    Ne_merged = Ne - Ne_simmed
    Ni_merged = Ni - Ni_simmed
    N_simmed = Ne_simmed + Ni_simmed
    if print_N:
        print(f"{Ne=}, {Ni=}, {N_simmed=}, {Ne_merged=}, {Ni_merged=}")
    
    n = COBA_AdEx_neuron()

    rates = lognormal(μ, σ, N_simmed) * Hz
    P = PoissonGroup(N_simmed, rates)
    Se = Synapses(P, n, on_pre="ge += we")
    Si = Synapses(P, n, on_pre="gi += wi")
    Se.connect("i < Ne_simmed")
    Si.connect("i >= Ne_simmed")

    PIe = PoissonInput(n, 'ge', Ne_merged, μₓ, we)
    PIi = PoissonInput(n, 'gi', Ni_merged, μₓ, wi);

    M = StateMonitor(n, ["V"], record=[0])
    S = SpikeMonitor(n)
    SP = SpikeMonitor(P)
    
    objs = [n, P, Se, Si, PIe, PIi, M, S, SP]
    return *objs, Network(objs)

%%time
*objs_m, net_m = Nto1_merged()

Ne=5200, Ni=1300, N_simmed=200, Ne_merged=5100, Ni_merged=1200
CPU times: total: 46.9 ms
Wall time: 376 ms

%%time
net_m.run(T, report='text')

CPU times: total: 93.8 ms
Wall time: 18.4 s

So, wall time of 23.4 seconds. That’s generating all the C++ etc files in ./output/, and then running.

We can get just the runtime by running the main ig (but that’s not useful irl ig, cause can’t change params w/o rebuilding (?)).

# ! output\\main.exe

(If run here, we get some IO errors).
But if run in sep terminal, we get no errors.
And.. “10 seconds simulated in < 1s”. Impressive!

Now with PIe and i included in net:
total runtime 52.2 s.
main.exe: 1.9 seconds.

2023-08-14 (after deleting cpp/):

1.6 wall time for Nto1_merged()
61 s wall time for run
0.898 s for time ./main.exe

61+1.6-0.898

61.702

2023-08-15 (after deleting cpp/, and w/ CPU in best performance mode (not battery saver)):

0.376 wall time for Nto1_merged()
18.4 s wall time for run
0.353 s for time ./main.exe

0.376+18.4-0.353

18.423

Ah, we can actually change params w/o rebuilding everything:
https://brian2.readthedocs.io/en/stable/examples/multiprocessing.02_using_standalone.html

you don’t need to recompile the entire project at each simulation. In the generated code, two consecutive simulations will only differ slightly (in this case only the tau parameter). The compiler will therefore only recompile the file that has changed and not the entire project

w/o PoissonInput merging¶

(Restarting nb and re-importing brian)

set_device('cpp_standalone', directory='cpp/2')

def Nto1_all_simmed(N = 6500):
    
    Ne = N * 4//5
    
    n = COBA_AdEx_neuron()
    
    rates = lognormal(μ, σ, N) * Hz
    P = PoissonGroup(N, rates)
    
    Se = Synapses(P, n, on_pre="ge += we")
    Si = Synapses(P, n, on_pre="gi += wi")
    Se.connect("i < Ne")
    Si.connect("i >= Ne")
    
    M = StateMonitor(n, ["V"], record=[0])
    S = SpikeMonitor(n)
    SP = SpikeMonitor(P)
    
    objs = [n, P, Se, Si, M, S, SP]
    return *objs, Network(objs)

%%time
*objs, net = Nto1_all_simmed()

CPU times: total: 93.8 ms
Wall time: 123 ms

%%time
net.run(T, report='text')

CPU times: total: 125 ms
Wall time: 21.7 s

New run: 42.2 s total.

main.exe: 28.9 s (!)

2023-08-14 (after deleting cpp/):

0.67 wall time for Nto1_all_simmed()
70 s wall time for run
21.43 s for time ./main.exe

70+0.67-21.43

49.24

2023-08-15 (after deleting cpp/, and w/ CPU in best performance mode (not battery saver)):

0.123 wall time for Nto1_all_simmed()
21.7 s wall time for run
6.50 s for time ./main.exe

0.123+21.7-6.5

15.323

Multiple runs¶

set_device('cpp_standalone', directory='cpp/3')

from time import time

for j, we in enumerate([8, 14, 20] * pS):
    print(f"Run {j+1} … ", end="")
    t0 = time()
    device.reinit()
    device.activate()
    *objs, net = Nto1_merged(print_N=False)
    net.run(T, report='text')
    print(f"{time() - t0:.1f} s")

Run 1 … 23.1 s
Run 2 … 18.7 s
Run 3 … 24.5 s

So, no caching speedup.

(Same conclusion when running https://brian2.readthedocs.io/en/stable/examples/multiprocessing.02_using_standalone.html: all all run_sim call take 12 à 15 seconds).

ok, finally, to reconfirm, https://brian2.readthedocs.io/en/stable/examples/multiprocessing.02_using_standalone.html with one proc.
- (cause that text literally says: “The compiler will therefore only recompile the file that has changed and not the entire project”)
- ye ok, there’s lil speedup:

6 s
6
2
..
2

Voltage-to-Wiring Sim

2023-08-02__speedtest_brian_standalone_AdEx_Nto1

Contents

2023-08-02__speedtest_brian_standalone_AdEx_Nto1¶

w/o PoissonInput merging¶

Multiple runs¶