Functions are the basic building blocks of a Kelp.Net neural network. Single functions are chained together within function stacks to create powerful and possibly complex network chains. There are four primary types of functions you need to know about, and their purposes, should be self-explanatory:
- Single-input functions
- Dual-input functions
- Multi-input functions
- Multi-output functions
Functions are also chained together when networks are loaded from disks.
Each function has a forward and backward method that you will be implementing when you create functions of your own:
public abstract NdArray[] Forward(params NdArray[] xs);
public virtual void Backward([CanBeNull] params NdArray[] ys){}