Error values are way high about drago HOT 5 CLOSED

It isn't, it's used strictly for calculations. Can I see how you're using the network?

from drago.

func main() {
    net := Drago.New(.1, 10000, []int{3, 5, 5, 1}, []Drago.Activator{new(Drago.Sigmoid), new(Drago.Sigmoid)})
    //net.Verbose = false
    samples := [][][]float64{
        {{20.937, 6.17, 22.792}, {6.42501}},
        {{20.375, 5.27, 20.5631}, {0.505808}},
        {{20.625, 5.27, 21.569}, {1.90032}},
        {{20.562, 5.27, 22.8776}, {6.343}},
        {{20.187, 5.27, 22.7772}, {6.58899}},
        //Lots more data
    }
    net.Learn(samples)
    fmt.Println(net.Predict([]float64{20.937, 6.17, 22.792}).At(0, 0))
}

from drago.

How many samples would you say you have, roughly? If you have an appreciable number, you should set the number of iterations far far lower, try 25 or something in that ballpark. Iterating 10000 times will lead to extreme overfitting, and negatively impact your results.

from drago.

Really wouldn't have expected that, typically you want a lot of iterations surely? Is this set up different to others I'm likely to have used? Being feed forward maybe? I have 50 samples.

from drago.

If you have too many iterations, you will overfit to the dataset that you trained on, and will get poor results when you are predicting values that were not in your original dataset. You can also experiment with different topologies / learning rates / numbers of iterations to see what will give you the best results. Also, getting more data samples would help, 50 is very few.

from drago.