func (e *engine) learn(req *request) {
r := *req
in := r[0].(*[]float64)
ideal := r[1].(*[]float64)
speed := r[2].(float64)
learn.Learn(e.Network, *in, *ideal, speed)
}
func MulriplyMain() {
network := neural.NewNetwork(2, []int{5, 5, 1})
network.RandomizeSynapses()
ch := make(chan float64, 300)
go func() {
tick := time.Tick(5 * time.Second)
acount := 1000
arr := make([]float64, acount)
index := 0
for {
select {
case v := <-ch:
arr[index] = v
index++
if index == len(arr) {
index = 0
}
case <-tick:
var sum float64 = 0
for _, val := range arr {
sum += val
}
//log.Println(index)
//log.Println(arr)
log.Printf("Avarege error: %f", sum/float64(acount))
}
}
}()
count := 100000000000
for i := 0; i < count; i++ {
test := []float64{rand.Float64(), rand.Float64()}
result := network.Calculate(test)[0]
ch <- math.Abs(test[0]*test[1] - result)
log.Printf("Error value: %f * %f = %f", test[0], test[1], test[0]*test[1])
learn.Learn(network, test, []float64{test[0] * test[1]}, 0.1)
}
}
func learnNetwork(n *neural.Network) {
samples := make([]*Sample, 0, 10)
samples = append(samples, loadSample("plus"))
samples = append(samples, loadSample("minus"))
samples = append(samples, loadSample("multiple"))
samples = append(samples, loadSample("divide"))
for i := 0; i < 10000; i++ {
for _, s := range samples {
learn.Learn(n, s.In, s.Out, speed)
}
}
}
func learnLangFile(n *neural.Network, path string, out []float64) {
//log.Println("Learning ", path)
sample := getSampleFromFile(path)
learn.Learn(n, sample, out, learningSpeed)
}