Practical Neural Network Recipies in C++

126

Chapter 7

at this guess, and the function value preserved as the best so far. The

temperature (standard deviation of the random perturbation) is

initialized to the user's starting temperature, and the reduction factor

is computed.

At the start of each pass through the temperature loop, the

improved flag is set to zero. When all iterations at that temperature

are complete, the perturbation center for the next temperature will be

updated only if that flag was set to one due to improvement. Other-

wise,

the same center will be used again.

Each iteration at a given temperature starts by calling randO

for a random number. We then call srandO