Computational Intelligence

Back-propagation Implementation 233

bp.mse /= bp.env.max_tra_pat;

if ((++bp.env.cur_gen) < bp.env.max_gen) // add error criterion later

{

bp.mse = 0.0; //clear mean squared error

bp_cur_state = BP_UPDATE_LEARNING_RATE;

}

else

{

bp_cur_state = BP_TRAINING_DONE;

}

}

In the two states, BP_UPDATE_LEARNING_RATE and BP_UPDATE_

MOMENTUM_RATE

, if a dynamic learning rate and/or momentum rate are

used, then the new learning rate and momentum rate are updated. In our implemen-

tation, rates are fixed. Therefore, these two state handler routines do nothing except

transition the current state to state

BP_UPDATE_MOMENTUM_RATE and state

BP_GET_PATTERN, respectively.

In the