math-optim/algo/de/jDE.go

// Copyright 2023 wanderer <a_mirre at utb dot cz>
// SPDX-License-Identifier: GPL-3.0-or-later

package de

import (
	"log"
	"os"
	"time"

	"git.dotya.ml/wanderer/math-optim/bench"
	"git.dotya.ml/wanderer/math-optim/stats"
)

// JDE is a holder for the settings of an instance of a self-adapting
// differential evolution (jDE) algorithm.
type JDE struct {
	// Generations denotes the number of generations the population evolves
	// for. Special value -1 disables limiting the number of generations.
	Generations int
	// BenchMinIters is the number of iterations the bench function will be re-run (for statistical purposes).
	BenchMinIters int
	// Dimensions to solve the problem for.
	Dimensions []int
	// F is the differential weight (mutation/weighting factor).
	F float64
	// CR is the crossover probability constant.
	CR float64
	// fVect holds the F values in a slice.
	fVect []float64
	// crVect holds the CR values in a slice.
	crVect []float64
	// MutationStrategy selects the mutation strategy, i.e. the variant of the
	// jDE algorithm (0..17), see mutationStrategies.go for more details.
	MutationStrategy int
	// AdptScheme is the parameter self-adaptation scheme (0..1).
	AdptScheme int
	// NP is the initial population size.
	NP int
	// BenchName is a name of the problem to optimise.
	BenchName string
	// ch is a channel for writing back computed results.
	ch chan []stats.Stats
	// initialised denotes the initialisation state of the struct.
	initialised bool
}

const (
	// jDEMinNP is the minimum size of the initial population for jDE.
	jDEMinNP = 4
	// fMin is the minimum allowed value of the differential weight.
	fMin = 0.5
	// fMax is the maximum allowed value of the differential weight.
	fMax = 2.0
	// crMin is the minimum allowed value of the crossover probability constant.
	crMin = 0.2
	// crMax is the maximum allowed value of the crossover probability constant.
	crMax = 0.9
)

// jDELogger declares and initialises a "custom" jDE logger.
var jDELogger = log.New(os.Stderr, " *** δ jDE:", log.Ldate|log.Ltime|log.Lshortfile)

// Init initialises the jDE algorithm, performs sanity checks on the inputs.
func (j *JDE) Init(generations, benchMinIters, mutStrategy, adptScheme, np int, f, cr float64, dimensions []int, bench string, ch chan []stats.Stats) {
	if j == nil {
		jDELogger.Fatalln("jDE needs to be initialised before calling RunjDE, exiting...")
	}

	// check input parameters.
	switch {
	case generations == 0:
		jDELogger.Fatalln("Generations cannot be 0, got", generations)

	case generations == -1:
		jDELogger.Println("Generations is '-1', disabling generation limits..")

	case benchMinIters < 1:
		jDELogger.Fatalln("Minimum bench iterations cannot be less than 1, got:", benchMinIters)

	case mutStrategy < 0 || mutStrategy > 17:
		jDELogger.Fatalln("Mutation strategy needs to be from the interval <0; 17>, got", mutStrategy)

	case adptScheme < 0 || adptScheme > 1:
		jDELogger.Fatalln("Parameter self-adaptation scheme needs to be from the interval <0; 1>, got", adptScheme)

	case np < jDEMinNP:
		jDELogger.Fatalf("NP cannot be less than %d, got: %d\n.", jDEMinNP, np)

	case f < fMin || f > fMax:
		jDELogger.Fatalf("F needs to be from the interval <%f;%f>, got: %f\n.", fMin, fMax, f)

	case cr < crMin || cr > crMax:
		jDELogger.Fatalf("CR needs to be from the interval <%f;>%f, got: %f\n.", crMin, crMax, cr)

	case len(dimensions) == 0:
		jDELogger.Fatalf("Dimensions cannot be empty, got: %+v\n", dimensions)

	case bench == "":
		jDELogger.Fatalln("Bench cannot be empty, got:", bench)
	}

	j.Generations = generations
	j.BenchMinIters = benchMinIters
	j.MutationStrategy = mutStrategy
	j.AdptScheme = adptScheme
	j.NP = np
	j.F = f
	j.CR = cr
	j.Dimensions = dimensions
	j.BenchName = bench
	j.ch = ch

	fV := make([]float64, np)
	crV := make([]float64, np)

	for i := 0; i < np; i++ {
		fV[i] = f
		crV[i] = cr
	}

	j.fVect = fV
	j.crVect = crV

	j.initialised = true
}

// InitAndRun initialises the jDE algorithm, performs sanity checks on the
// inputs and calls the Run method.
func (j *JDE) InitAndRun(generations, benchMinIters, mutStrategy, adptScheme, np int, f, cr float64, dimensions []int, bench string, ch chan []stats.Stats) {
	if j == nil {
		jDELogger.Fatalln("jDE is nil, NewjDE() needs to be called first. exiting...")
	}

	j.Init(generations, benchMinIters, mutStrategy, adptScheme, np, f, cr, dimensions, bench, ch)

	j.Run()
}

// Run self-adapting differential evolution algorithm.
func (j *JDE) Run() {
	if j == nil {
		jDELogger.Fatalln("jDE is nil, NewjDE() needs to be called first. exiting...")
	}

	if !j.initialised {
		jDELogger.Fatalln("jDE needs to be initialised before calling Run(), exiting...")
	}

	// run evolve for for all dimensions.
	for _, dim := range j.Dimensions {
		maxFES := bench.GetGAMaxFES(dim)

		// create a population with known params.
		pop := newPopulation(j.BenchName, j.NP, dim)

		// set population seed.
		pop.Seed = uint64(time.Now().UnixNano())
		// initialise the population.
		pop.Init()

		j.evolve(maxFES, pop)
	}
}

// evolve evolves a population by running the jDE (self-adapting Differential
// Evolution) algorithm on the passed population until termination conditions
// are met.
func (j *JDE) evolve(maxFES int, pop *Population) {}

// NewjDE returns a pointer to a new, uninitialised jDE instance.
func NewjDE() *JDE {
	return &JDE{}
}

// LogPrintln wraps the jDE logger's Println func.
func LogPrintln(v ...any) {
	jDELogger.Println(v...)
}

// LogPrintf wraps the jDE logger's Printf func.
func LogPrintf(s string, v ...any) {
	jDELogger.Printf(s, v...)
}

// LogFatalln wraps the jDE logger's Fatalln func.
func LogFatalln(s string) {
	jDELogger.Fatalln(s)
}

// LogFatalf wraps the jDE logger's Fatalf func.
func LogFatalf(s string, v ...any) {
	jDELogger.Fatalf(s, v...)
}