On the effects of topology and node distribution on learning over complex adaptive networks