Does hill climbing have any other problems we should worry about? >> Suppose we start at the left edge of the graph, using small steps we gradually climb higher until we get to the shoulder here. However, if we stop when we no longer see a way to improve we can get stuck at this shoulder. >> So you mean we get to this point and if the step size is small enough we never see the sudden increase in gradient here? >> Yup. For flat areas how do we know which way to travel? If we just choose a direction randomly we can wander on this plateau for a while. The code needs to return a result at some point. So how many times in a row should we allow a zero improvement in score before we stop? We could just keep going in the direction of the last positive gradient. But with a small enough step size, we can easily think we are not improving and the algorithm should stop.