Practical Neural Network Recipies in C++

270

Chapter 16

when it may not be necessary? Time spent taking care of this detail

can be paid back handsomely in improved performance and training

time.

Circular Discontinuity

Sometimes the variables we present to neural networks are fundamen-

tally circular. The position of a rotating shaft is an obvious example.

The azimuth angle at which we are viewing a solid object is another.

Even color, measured on many popular scales, is circular. These vari-

ables introduce a special problem due to the fact that they have a

discontinuity. If

we

define the position of a rotating object as being an

angle measured from 0 to 360 degrees, we have a seriou ...