func ExampleModel() {
gp.SetSeed(1)
pset := gp.CreatePrimSet(1, "x")
pset.Add(num.Add, num.Sub, num.Mul, num.Div, num.Neg, num.V(0), num.V(1))
problem := gp.Model{
PrimitiveSet: pset,
Generator: gp.GenFull(pset, 1, 3),
PopSize: 500,
Fitness: getFitness,
Offspring: gp.Tournament(3),
Mutate: gp.MutUniform(gp.GenGrow(pset, 0, 2)),
MutateProb: 0.2,
Crossover: gp.CxOnePoint(),
CrossoverProb: 0.5,
Threads: 1,
}
logger := &stats.Logger{MaxGen: 20, TargetFitness: 0.99, PrintStats: true}
problem.Run(logger)
// Output:
// set random seed: 1
// Gen Evals FitMax FitAvg FitStd SizeAvg SizeMax DepthAvg DepthMax
// 0 500 0.12 0.025 0.014 6.85 15 1.96 3
// 1 299 0.33 0.0344 0.0204 6.33 27 1.93 6
// 2 286 0.663 0.0469 0.0448 6.26 27 1.9 7
// 3 265 0.663 0.0598 0.0683 6.58 34 2.06 9
// 4 280 0.663 0.0772 0.088 7.51 39 2.39 9
// 5 291 0.663 0.0918 0.1 8.92 32 2.82 8
// 6 302 0.663 0.117 0.133 10.3 35 3.2 10
// 7 294 1 0.152 0.17 11.1 35 3.48 10
// ** SUCCESS **
}
func Example_gp() {
// create initial population
gp.SetSeed(1)
pset := gp.CreatePrimSet(1, "x")
pset.Add(num.Add, num.Sub, num.Mul, num.Div, num.Neg, num.V(0), num.V(1))
generator := gp.GenFull(pset, 1, 3)
pop, evals := gp.CreatePopulation(500, generator).Evaluate(eval{}, 1)
best := pop.Best()
fmt.Printf("gen=%d evals=%d fit=%.4f\n", 0, evals, best.Fitness)
// setup genetic variations
tournament := gp.Tournament(3)
mutate := gp.MutUniform(gp.GenGrow(pset, 0, 2))
crossover := gp.CxOnePoint()
// loop till reach target fitness or exceed no. of generations
for gen := 1; gen <= 40 && best.Fitness < 1; gen++ {
offspring := tournament.Select(pop, len(pop))
pop, evals = gp.VarAnd(offspring, crossover, mutate, 0.5, 0.2).Evaluate(eval{}, 1)
best = pop.Best()
fmt.Printf("gen=%d evals=%d fit=%.4f\n", gen, evals, best.Fitness)
}
fmt.Println(best.Code.Format())
// Output:
// set random seed: 1
// gen=0 evals=500 fit=0.1203
// gen=1 evals=299 fit=0.3299
// gen=2 evals=286 fit=0.6633
// gen=3 evals=265 fit=0.6633
// gen=4 evals=280 fit=0.6633
// gen=5 evals=291 fit=0.6633
// gen=6 evals=302 fit=0.6633
// gen=7 evals=294 fit=1.0000
// (x + (((x / 1) - ((x / 1) * -(((x * x) + x)))) * (1 * x)))
}
func getStats(t *testing.T, gen int) *Stats {
pset := gp.CreatePrimSet(1, "x")
pset.Add(num.Add, num.Sub, num.Mul, num.Div, num.Neg, num.V(0), num.V(1))
pop := gp.CreatePopulation(1000, gp.GenFull(pset, 1, 3))
for i := range pop {
pop[i].Fitness = rand.Float64()
}
s := Create(pop, gen, len(pop))
t.Log(s)
return s
}
// calc least squares difference and return as normalised fitness from 0->1
func (e eval) GetFitness(code gp.Expr) (float64, bool) {
diff := 0.0
for x := -1.0; x <= 1.0; x += 0.1 {
val := float64(code.Eval(num.V(x)).(num.V))
fun := x*x*x*x + x*x*x + x*x + x
diff += (val - fun) * (val - fun)
}
return 1.0 / (1.0 + diff), true
}
// returns function to calc least squares difference and return as normalised fitness from 0->1
func fitnessFunc(trainSet []Point) func(gp.Expr) (float64, bool) {
return func(code gp.Expr) (float64, bool) {
diff := 0.0
for _, pt := range trainSet {
val := float64(code.Eval(num.V(pt.x)).(num.V))
diff += (val - pt.y) * (val - pt.y)
}
return 1.0 / (1.0 + diff), true
}
}
// function to plot best individual
func plotBest(trainSet []Point) func(gp.Population) stats.Plot {
return func(pop gp.Population) stats.Plot {
plot := stats.NewPlot("Best", len(trainSet))
plot.Color = "#ff0000"
code := pop.Best().Code
for i, pt := range trainSet {
plot.Data[i][0] = pt.x
plot.Data[i][1] = float64(code.Eval(num.V(pt.x)).(num.V))
}
return plot
}
}
// function to generate the random constant generator function
func ercGen(start, end int) func() num.V {
fmt.Println("generate random constants in range", start, "to", end)
return func() num.V {
return num.V(start + rand.Intn(end-start+1))
}
}