Implement relu

[kasey-]

Hello,

I started to implement the relu function for the genann library on a fork under my name (https://github.com/kasey-/genann) before sending you a PR:

double inline genann_act_relu(const struct genann *ann unused, double a) {
    return (a > 0.0) ? a : 0.0;
}

But I am a bit lost in the way you compute the back propagation of the neural network. The derivative of relu formula is trivial (a > 0.0) ? 1.0 : 0.0 But I cannot understand were I should plug-it inside your formula as I do not understand how do you compute your back propagation. Did you implemented only the derivate of the sigmoid ?

[ilia3101]

I would like ReLU in genann too, but the back propagation is also something I don't understand here :(

[msrdinesh]

Yes, In the code derivate of sigmoid "ddxσ(x)=σ(x)(1−σ(x))" is only implemented. I think we have to write a generic function of derivatives so that, we can add other activation functions like tanh and Relu check the code here https://github.com/kasey-/genann/blob/27c4c4288728791def0c5fd175c1c3999057ad9d/genann.c#L335 If you agree I can also work on this.

[AnnieJohnson25]

Hi everyone, To actually implement Relu and linear what exactly should we be looking at other than the back propagation derivative? Would it also require any additional functions like how sigmoid has genann_act_sigmoid_cached and genann_init_sigmoid_lookup? Any advice...

[doug65536]

Hi everyone, To actually implement Relu and linear what exactly should we be looking at other than the back propagation derivative? [snip]

It's right there at the top of this bug report thread, double inline genann_act_relu. (To be clear: It still doesn't work using that, because the derivative is hardcoded to sigmoid).