Skip to content

GitLab

Commit e5624bd7 authored by K-H-Ismail's avatar K-H-Ismail Committed by Gauthier Quesnel
Browse files

benchmark: add LIF and Izhikevich benchmark

parent b80fe629
Hide whitespace changes
Inline Side-by-side
lib/benchmark/benchmark2.cpp 0 → 100644
View file @ e5624bd7
// Copyright (c) 2020 INRA Distributed under the Boost Software License,
// Version 1.0. (See accompanying file LICENSE_1_0.txt or copy at
// http://www.boost.org/LICENSE_1_0.txt)
#include <irritator/core.hpp>
#include <boost/ut.hpp>
#include <fmt/format.h>
#include <cstdio>
#include <fstream>
#include <hayai/hayai.hpp>
using namespace std;
struct file_output
{
file_output(const char* file_path) noexcept
: os(std::fopen(file_path, "w"))
{}
~file_output() noexcept
{
if (os)
std::fclose(os);
}
std::FILE* os = nullptr;
};
void
file_output_initialize(const irt::observer& obs, const irt::time /*t*/) noexcept
{
if (!obs.user_data)
return;
auto* output = reinterpret_cast<file_output*>(obs.user_data);
fmt::print(output->os, "t,{}\n", obs.name.c_str());
}
void
file_output_observe(const irt::observer& obs,
const irt::time t,
const irt::message& msg) noexcept
{
if (!obs.user_data)
return;
auto* output = reinterpret_cast<file_output*>(obs.user_data);
fmt::print(output->os, "{},{}\n", t, msg.to_real_64(0));
}
struct neuron {
irt::dynamics_id sum;
irt::dynamics_id prod;
irt::dynamics_id integrator;
irt::dynamics_id quantifier;
irt::dynamics_id constant;
irt::dynamics_id cross;
irt::dynamics_id constant_cross;
};
struct neuron
make_neuron(irt::simulation* sim, long unsigned int i, double quantum) noexcept
{
using namespace boost::ut;
double tau_lif = 10;
double Vr_lif = 0.0;
double Vt_lif = 10.0;
/*double ref_lif = 0.5*1e-3;
double sigma_lif = 0.02;*/
auto& sum_lif = sim->adder_2_models.alloc();
auto& prod_lif = sim->adder_2_models.alloc();
auto& integrator_lif = sim->integrator_models.alloc();
auto& quantifier_lif = sim->quantifier_models.alloc();
auto& constant_lif = sim->constant_models.alloc();
auto& constant_cross_lif = sim->constant_models.alloc();
auto& cross_lif = sim->cross_models.alloc();
sum_lif.default_input_coeffs[0] = -1.0;
sum_lif.default_input_coeffs[1] = 20.0;
prod_lif.default_input_coeffs[0] = 1.0/tau_lif;
prod_lif.default_input_coeffs[1] = 0.0;
constant_lif.default_value = 1.0;
constant_cross_lif.default_value = Vr_lif;
integrator_lif.default_current_value = 0.0;
quantifier_lif.default_adapt_state =
irt::quantifier::adapt_state::possible;
quantifier_lif.default_zero_init_offset = true;
quantifier_lif.default_step_size = quantum;
quantifier_lif.default_past_length = 3;
cross_lif.default_threshold = Vt_lif;
sim->alloc(sum_lif, sim->adder_2_models.get_id(sum_lif));
sim->alloc(prod_lif, sim->adder_2_models.get_id(prod_lif));
sim->alloc(integrator_lif, sim->integrator_models.get_id(integrator_lif));
sim->alloc(quantifier_lif, sim->quantifier_models.get_id(quantifier_lif));
sim->alloc(constant_lif, sim->constant_models.get_id(constant_lif));
sim->alloc(cross_lif, sim->cross_models.get_id(cross_lif));
sim->alloc(constant_cross_lif, sim->constant_models.get_id(constant_cross_lif));
struct neuron neuron_model = {sim->adder_2_models.get_id(sum_lif),
sim->adder_2_models.get_id(prod_lif),
sim->integrator_models.get_id(integrator_lif),
sim->quantifier_models.get_id(quantifier_lif),
sim->constant_models.get_id(constant_lif),
sim->cross_models.get_id(cross_lif),
sim->constant_models.get_id(constant_cross_lif),
};
// Connections
expect(sim->connect(quantifier_lif.y[0], integrator_lif.x[0]) ==
irt::status::success);
expect(sim->connect(prod_lif.y[0], integrator_lif.x[1]) ==
irt::status::success);
expect(sim->connect(cross_lif.y[0], integrator_lif.x[2]) ==
irt::status::success);
expect(sim->connect(cross_lif.y[0], quantifier_lif.x[0]) ==
irt::status::success);
expect(sim->connect(cross_lif.y[0], sum_lif.x[0]) ==
irt::status::success);
expect(sim->connect(integrator_lif.y[0],cross_lif.x[0]) ==
irt::status::success);
expect(sim->connect(integrator_lif.y[0],cross_lif.x[2]) ==
irt::status::success);
expect(sim->connect(constant_cross_lif.y[0],cross_lif.x[1]) ==
irt::status::success);
expect(sim->connect(constant_lif.y[0], sum_lif.x[1]) ==
irt::status::success);
expect(sim->connect(sum_lif.y[0],prod_lif.x[0]) ==
irt::status::success);
expect(sim->connect(constant_lif.y[0],prod_lif.x[1]) ==
irt::status::success);
return neuron_model;
}
void lif_benchmark(double simulation_duration, double quantum)
{
using namespace boost::ut;
irt::simulation sim;
long unsigned int N = 1;
expect(irt::is_success(sim.init(2600lu, 40000lu)));
// Neurons
std::vector<struct neuron> first_layer_neurons;
for (long unsigned int i = 0 ; i < N; i++) {
struct neuron neuron_model = make_neuron(&sim,i,quantum);
first_layer_neurons.emplace_back(neuron_model);
}
irt::time t = 0.0;
std::string file_name = "output_lif_sd_"+
std::to_string(simulation_duration)+
"_q_"+std::to_string(quantum)+
".csv";
std::FILE* os = std::fopen(file_name.c_str(), "w");
!expect(os != nullptr);
std::string s = "t,";
for (long unsigned int i = 0; i < N; i++)
{
s = s + "spikes"
+ ","+ "v"
+ ",";
}
fmt::print(os, s + "\n");
expect(irt::status::success == sim.initialize(t));
do {
irt::status st = sim.run(t);
expect(st == irt::status::success);
std::string s = std::to_string(t)+",";
for (long unsigned int i = 0; i < N; i++)
{
s = s + std::to_string(sim.cross_models.get(first_layer_neurons[i].cross).event)
+ ",";
s = s + std::to_string(sim.integrator_models.get(first_layer_neurons[i].integrator).last_output_value)
+ ",";
}
fmt::print(os, s + "\n");
} while (t < simulation_duration);
std::fclose(os);
}
void izhikevich_benchmark(double simulation_duration, double quantum, double a, double b, double c, double d)
{
using namespace boost::ut;
irt::simulation sim;
expect(irt::is_success(sim.init(1000lu, 1000lu)));
expect(sim.constant_models.can_alloc(3));
expect(sim.adder_2_models.can_alloc(3));
expect(sim.adder_4_models.can_alloc(1));
expect(sim.mult_2_models.can_alloc(1));
expect(sim.integrator_models.can_alloc(2));
expect(sim.quantifier_models.can_alloc(2));
expect(sim.cross_models.can_alloc(2));
auto& constant = sim.constant_models.alloc();
auto& constant2 = sim.constant_models.alloc();
auto& constant3 = sim.constant_models.alloc();
auto& sum_a = sim.adder_2_models.alloc();
auto& sum_b = sim.adder_2_models.alloc();
auto& sum_c = sim.adder_4_models.alloc();
auto& sum_d = sim.adder_2_models.alloc();
auto& product = sim.mult_2_models.alloc();
auto& integrator_a = sim.integrator_models.alloc();
auto& integrator_b = sim.integrator_models.alloc();
auto& quantifier_a = sim.quantifier_models.alloc();
auto& quantifier_b = sim.quantifier_models.alloc();
auto& cross = sim.cross_models.alloc();
auto& cross2 = sim.cross_models.alloc();
double I = 10.0;
double vt = 30.0;
constant.default_value = 1.0;
constant2.default_value = c;
constant3.default_value = I;
cross.default_threshold = vt;
cross2.default_threshold = vt;
integrator_a.default_current_value = 0.0;
quantifier_a.default_adapt_state =
irt::quantifier::adapt_state::possible;
quantifier_a.default_zero_init_offset = true;
quantifier_a.default_step_size = quantum;
quantifier_a.default_past_length = 3;
integrator_b.default_current_value = 0.0;
quantifier_b.default_adapt_state =
irt::quantifier::adapt_state::possible;
quantifier_b.default_zero_init_offset = true;
quantifier_b.default_step_size = quantum;
quantifier_b.default_past_length = 3;
product.default_input_coeffs[0] = 1.0;
product.default_input_coeffs[1] = 1.0;
sum_a.default_input_coeffs[0] = 1.0;
sum_a.default_input_coeffs[1] = -1.0;
sum_b.default_input_coeffs[0] = -a;
sum_b.default_input_coeffs[1] = a * b;
sum_c.default_input_coeffs[0] = 0.04;
sum_c.default_input_coeffs[1] = 5.0;
sum_c.default_input_coeffs[2] = 140.0;
sum_c.default_input_coeffs[3] = 1.0;
sum_d.default_input_coeffs[0] = 1.0;
sum_d.default_input_coeffs[1] = d;
expect(sim.models.can_alloc(14));
!expect(irt::is_success(
sim.alloc(constant3, sim.constant_models.get_id(constant3), "tfun")));
!expect(irt::is_success(
sim.alloc(constant, sim.constant_models.get_id(constant), "1.0")));
!expect(irt::is_success(sim.alloc(
constant2, sim.constant_models.get_id(constant2), "-56.0")));
!expect(irt::is_success(
sim.alloc(sum_a, sim.adder_2_models.get_id(sum_a), "sum_a")));
!expect(irt::is_success(
sim.alloc(sum_b, sim.adder_2_models.get_id(sum_b), "sum_b")));
!expect(irt::is_success(
sim.alloc(sum_c, sim.adder_4_models.get_id(sum_c), "sum_c")));
!expect(irt::is_success(
sim.alloc(sum_d, sim.adder_2_models.get_id(sum_d), "sum_d")));
!expect(irt::is_success(
sim.alloc(product, sim.mult_2_models.get_id(product), "prod")));
!expect(irt::is_success(sim.alloc(
integrator_a, sim.integrator_models.get_id(integrator_a), "int_a")));
!expect(irt::is_success(sim.alloc(
integrator_b, sim.integrator_models.get_id(integrator_b), "int_b")));
!expect(irt::is_success(sim.alloc(
quantifier_a, sim.quantifier_models.get_id(quantifier_a), "qua_a")));
!expect(irt::is_success(sim.alloc(
quantifier_b, sim.quantifier_models.get_id(quantifier_b), "qua_b")));
!expect(irt::is_success(
sim.alloc(cross, sim.cross_models.get_id(cross), "cross")));
!expect(irt::is_success(
sim.alloc(cross2, sim.cross_models.get_id(cross2), "cross2")));
!expect(sim.models.size() == 14_ul);
expect(sim.connect(integrator_a.y[0], cross.x[0]) ==
irt::status::success);
expect(sim.connect(constant2.y[0], cross.x[1]) == irt::status::success);
expect(sim.connect(integrator_a.y[0], cross.x[2]) ==
irt::status::success);
expect(sim.connect(cross.y[0], quantifier_a.x[0]) ==
irt::status::success);
expect(sim.connect(cross.y[0], product.x[0]) == irt::status::success);
expect(sim.connect(cross.y[0], product.x[1]) == irt::status::success);
expect(sim.connect(product.y[0], sum_c.x[0]) == irt::status::success);
expect(sim.connect(cross.y[0], sum_c.x[1]) == irt::status::success);
expect(sim.connect(cross.y[0], sum_b.x[1]) == irt::status::success);
expect(sim.connect(constant.y[0], sum_c.x[2]) == irt::status::success);
expect(sim.connect(constant3.y[0], sum_c.x[3]) == irt::status::success);
expect(sim.connect(sum_c.y[0], sum_a.x[0]) == irt::status::success);
expect(sim.connect(integrator_b.y[0], sum_a.x[1]) ==
irt::status::success);
expect(sim.connect(cross2.y[0], sum_a.x[1]) == irt::status::success);
expect(sim.connect(sum_a.y[0], integrator_a.x[1]) ==
irt::status::success);
expect(sim.connect(cross.y[0], integrator_a.x[2]) ==
irt::status::success);
expect(sim.connect(quantifier_a.y[0], integrator_a.x[0]) ==
irt::status::success);
expect(sim.connect(cross2.y[0], quantifier_b.x[0]) ==
irt::status::success);
expect(sim.connect(cross2.y[0], sum_b.x[0]) == irt::status::success);
expect(sim.connect(quantifier_b.y[0], integrator_b.x[0]) ==
irt::status::success);
expect(sim.connect(sum_b.y[0], integrator_b.x[1]) ==
irt::status::success);
expect(sim.connect(cross2.y[0], integrator_b.x[2]) ==
irt::status::success);
expect(sim.connect(integrator_a.y[0], cross2.x[0]) ==
irt::status::success);
expect(sim.connect(integrator_b.y[0], cross2.x[2]) ==
irt::status::success);
expect(sim.connect(sum_d.y[0], cross2.x[1]) == irt::status::success);
expect(sim.connect(integrator_b.y[0], sum_d.x[0]) ==
irt::status::success);
expect(sim.connect(constant.y[0], sum_d.x[1]) == irt::status::success);
std::string file_name = "output_izhikevitch_a_sd_"+
std::to_string(simulation_duration)+
"_q_"+std::to_string(quantum)+
"_a_"+std::to_string(a)+
"_b_"+std::to_string(b)+
"_c_"+std::to_string(c)+
"_d_"+std::to_string(d)+
".csv";
file_output fo_a(file_name.c_str());
expect(fo_a.os != nullptr);
auto& obs_a = sim.observers.alloc(0.01,
"A",
static_cast<void*>(&fo_a),
&file_output_initialize,
&file_output_observe,
nullptr);
file_name = "output_izhikevitch_b_sd_"+
std::to_string(simulation_duration)+
"_q_"+std::to_string(quantum)+
"_a_"+std::to_string(a)+
"_b_"+std::to_string(b)+
"_c_"+std::to_string(c)+
"_d_"+std::to_string(d)+
".csv";
file_output fo_b(file_name.c_str());
expect(fo_b.os != nullptr);
auto& obs_b = sim.observers.alloc(0.01,
"B",
static_cast<void*>(&fo_b),
&file_output_initialize,
&file_output_observe,
nullptr);
sim.observe(sim.models.get(integrator_a.id), obs_a);
sim.observe(sim.models.get(integrator_b.id), obs_b);
irt::time t = 0.0;
expect(irt::status::success == sim.initialize(t));
!expect(sim.sched.size() == 14_ul);
do {
irt::status st = sim.run(t);
expect(st == irt::status::success);
} while (t < simulation_duration);
};
BENCHMARK_P(LIF, 1, 10, 1,( double simulation_duration, double quantum))
{
lif_benchmark(simulation_duration,quantum);
}
BENCHMARK_P(Izhikevich, Type, 1, 1,( double simulation_duration, double quantum, double a, double b, double c, double d))
{
izhikevich_benchmark(simulation_duration,quantum,a,b,c,d);
}
BENCHMARK_P_INSTANCE(LIF, 1, (30,1e-2));
// Regular spiking (RS)
BENCHMARK_P_INSTANCE(Izhikevich, Type, (1000,1e-2,0.02,0.2,-65.0,8.0));
// Intrinsical bursting (IB)
BENCHMARK_P_INSTANCE(Izhikevich, Type, (1000,1e-2,0.02,0.2,-55.0,4.0));
// Chattering spiking (CH)
BENCHMARK_P_INSTANCE(Izhikevich, Type, (1000,1e-2,0.02,0.2,-50.0,2.0));
// Fast spiking (FS)
BENCHMARK_P_INSTANCE(Izhikevich, Type, (1000,1e-2,0.1,0.2,-65.0,2.0));
// Thalamo-Cortical (TC)
BENCHMARK_P_INSTANCE(Izhikevich, Type, (1000,1e-2,0.02,0.25,-65.0,0.05));
// Rezonator (RZ)
BENCHMARK_P_INSTANCE(Izhikevich, Type, (1000,1e-2,0.1,0.26,-65.0,2.0));
// Low-threshold spiking (LTS)
BENCHMARK_P_INSTANCE(Izhikevich, Type, (1000,1e-2,0.02,0.25,-65.0,2.0));
// Problematic (P)
BENCHMARK_P_INSTANCE(Izhikevich, Type, (1000,1e-2,0.2,2,-56.0,-16.0));
int
main()
{
hayai::ConsoleOutputter consoleOutputter;
hayai::Benchmarker::AddOutputter(consoleOutputter);
hayai::Benchmarker::RunAllTests();
}
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment