Hi, thanks for the kind words!

The sigmoid has an output between 0 and 1, that might not be what you want depending on your regression problem. I think you can try to replicate this architecture here: https://www.google.com/amp/s/datascienceplus.com/keras-regression-based-neural-networks/amp/

You’ll need to implement two extra activations, relu activation (y=max(0,x)) and linear activation (y=x).

Hit me up if you need more help!