signalgp-lite

• Free software: MIT license
• Documentation: https://signalgp-lite.readthedocs.io
• header-only, namespace-encapsulated software

A genetic programming implementation designed for large-scale artificial life applications. Organized as a header-only C++ library. Inspired by Alex Lalejini's SignalGP.

Quick Start

This "hello world" example throws together * a custom hardware peripheral to manage greeting information, * a custom operation to print a greeting, and * generation of a random program, * execution of that random program on a virtual multi-core CPU.

say-hello.cpp: ```cpp

include

include

include

include "Empirical/include/emp/math/Random.hpp"

include "sgpl/algorithm/execute_cpu.hpp"

include "sgpl/spec/Spec.hpp"

include "sgpl/hardware/Cpu.hpp"

include "sgpl/library/OpLibraryCoupler.hpp"

include "sgpl/library/prefab/ControlFlowOpLibrary.hpp"

include "sgpl/program/Program.hpp"

emp::Random rng;

// custom hardware peripheral, can be written to or read from during execution struct Peripheral { sizet greetcount{}; std::string name; };

// custom CPU operation struct SayHello {

template static void run( sgpl::Core&, const sgpl::Instruction&, const sgpl::Program&, typename Spec::peripheralt& peripheral ) { std::cout << "for the " << peripheral.greetcount++ << "th time... "; std::cout << "hello there " << peripheral.name << '\n'; }

static std::string name() { return "SayHello"; }

static size_t prevalence() { return 1; }

};

// extends prefab ControlFlowOpLibrary with SayHello operation using library_t = sgpl::OpLibraryCoupler;

// custom compile-time configurator type using spect = sgpl::Spec<libraryt, Peripheral>;

int main() {

sgpl::Cpu cpu; Peripheral peripheral; peripheral.name = "Grace Hopper";

sgpl::Program program{ 100 }; // randomly generated, 100 instructions

cpu.InitializeAnchors( program ); // load program onto CPU

// generate random signals to launch available virtual cores while ( cpu.TryLaunchCore( emp::BitSet<64>(rng) ) ) ;

// execute up to one thousand instructions sgpl::executecpu<spect>( std::kilo::num, cpu, program, peripheral );

} ```

compile: bash g++ --std=c++17 -Iinclude/ -Ithird-party/ say-hello.cpp -o say-hello.out

run: bash ./say-hello.out

Benchmarks

signalgp-lite provides several-times speedup over the current "vanilla" SignalGP implementation.

Speedup of mean instruction execution time provided by signalgp-lite compared to vanilla SignalGP. Speedup is measured for random programs generated from different subsets of instructions ("libraries") over different-size populations of virtual CPUs ("num agents").

For randomly-generated programs composed of arbitrary instructions, signalgp-lite approaches a virtual instruction execution rate of around 10Mhz on a 3.5Ghz processor. Virtual nop instructions execute at rate of around 200Mhz.

Wall clock timings of twenty randomly-generated programs composed of instructions from different libraries.

Timings for nop and arithmetic libraries report the mean time to execute sixteen instructions on one core. Timings for complete and sans_regulation libraries report timings for executing sixteen instructions, one each across sixteen virtual threads. (sans_regulation refers to the complete library with tag-matching regulation disabled.)

These results are associated with commit c10ed70, measured at 1602292830 seconds since epoch. Details on the machine used to perform these benchmarks are available via Open Science Framework, e.g., https://osf.io/hu8m2/. mimalloc memory allocator.

Microbenchmarks are performed, graphed, and uploaded as part of the project's CI build, so check the project's OSF page for up-to-the-minute profiling information!

Citing

If signalgp-lite contributes to a scientific publication, please cite it as

Moreno, M. A., Rodriguez Papa, S., & Ofria, C. (2021). SignalGP-Lite: Event Driven Genetic Programming Library for Large-Scale Artificial Life Applications. arXiv preprint arXiv:2108.00382.

bibtex @misc{moreno2021signalgp, doi = {10.48550/ARXIV.2108.00382}, url = {https://arxiv.org/abs/2108.00382}, author = {Moreno, Matthew Andres and Rodriguez Papa, Santiago and Lalejini, Alexander and Ofria, Charles}, keywords = {Neural and Evolutionary Computing (cs.NE), FOS: Computer and information sciences, FOS: Computer and information sciences}, title = {SignalGP-Lite: Event Driven Genetic Programming Library for Large-Scale Artificial Life Applications}, publisher = {arXiv}, year = {2021}, copyright = {arXiv.org perpetual, non-exclusive license} }

Consider also citing Empirical. And don't forget to leave a star on GitHub!

Credits

This library draws heavily on Alex Lalejini's work with SignalGP.

This package was created with Cookiecutter and the devosoft/cookiecutter-empirical-project project template.

This package uses Empirical, a library of tools for scientific software development, with emphasis on also being able to build web interfaces using Emscripten.