The following things should be implemented in the ffnn class to use it as a compositional pattern producing network (cppn) for the indirect ctrnn encoding (aka hyperbrain):
[x] Investigate and tune the speed of the step/predict method, since this is a major bottleneck for hyperbrain (very important)
[x] Add a linear activation function to the output layer to get an output value range [-inf; +inf] of the ffnn (in order to encode weights)
bjuergens
commented
3 years ago
The following things should be implemented in the ffnn class to use it as a compositional pattern producing network (cppn) for the indirect ctrnn encoding (aka hyperbrain):