func q9cc() experiment {
exp := experiment{NRuns: 1000, NPoints: 100}
pla := pla.NewPLA()
chanDisagreement := make(chan float64, exp.NRuns)
chanIterations := make(chan int, exp.NRuns)
for run := 0; run < exp.NRuns; run++ {
go func() {
pla.N = exp.NPoints
pla.Initialize()
iterations := pla.Converge()
chanDisagreement <- pla.Disagreement()
chanIterations <- iterations
}()
}
for i := 0; i < exp.NRuns; i++ {
exp.SumOfDisagreement += <-chanDisagreement
}
for i := 0; i < exp.NRuns; i++ {
exp.SumOfIterations += <-chanIterations
}
exp.print()
return exp
}
// Take N = 10. How many iterations does it take on average for the PLA to
// converge for N = 10 training points?
func q7() experiment {
exp := experiment{NRuns: 1000, NPoints: 10}
pla := pla.NewPLA()
for run := 0; run < exp.NRuns; run++ {
pla.Initialize()
iterations := pla.Converge()
exp.SumOfDisagreement += pla.Disagreement()
exp.SumOfIterations += iterations
}
exp.print()
return exp
}
func q7() {
runs := 1000 // number of times we repeat the experiment
linreg := linreg.NewLinearRegression()
linreg.N = 10
iterations := 0
for run := 0; run < runs; run++ {
linreg.Initialize()
linreg.Learn()
pla := pla.NewPLA()
pla.Initialize()
for i := 0; i < len(pla.Wn); i++ {
pla.Wn[i] = linreg.Wn[i]
}
iterations += pla.Converge()
}
avgIterations := float64(iterations) / float64(runs)
fmt.Printf("average for PLA to converge for N = %v with initial weight computed by linear regression is %4.2f\n", linreg.N, avgIterations)
}
