Hill Climbing/Descent Sample Clauses

Hill Climbing/Descent. Hill Climbing is a method where a point in the design space is chosen, either randomly or according to some engineering intuition, and then then the space is searched by repeatedly sampling (simulating) a neighbouring point in the design space and testing whether this point has a higher value than the current point and jumping to it if it does. This process continues until there are no neighbouring points with higher values and then the current design is returned as the maximum. Hill Descent follows the same procedure but selects neighbours with lower values instead. Both of these methods can find a maximum with fewer simulations than either a random or exhaustive search, but they have a weakness that in a system which contains both local and global maxima, they will not always find the global maxima. There is also the steepest ascent/descent versions of this method, where in- stead of picking a single neighbour and determining if this is higher/lower, all surrounding neighbour designs are examined and the one with the high- est/lowest value is chosen. This has the potential to arrive at a maxima/min- ima earlier than the normal method, but still suffers from the same weakness of becoming trapped. Here the DSE driver needs to know the range of parameters and the resolution of each. This is a closed loop method and so it requires that the simulation results in either a single value or a comparable set of results.