Some Benchmarking Results

So I graphed what I did from benchmarking, and I got some very interesting graphs. One of the things I need to determine before the I finish designing the improvement to my algorithm, is something I call the "q-decay" factor. In this spirit of divide and conquer, the algorithm takes considerably large chunks and splits them into smaller chunks. The size of these chunks are determined by the q value. So what would be a good rate of decay? Using the dataset I was given, I tried out different rates of decay: A factor of 10, a factor of 5 and a factor of 2. Below is the graph for the decay according to a factor of 2:

Pretty, isn't it? I really liked how gently the slopes seemed to decay. This is very favorible. In order to give you an idea of what is NOT favorible, I think I should show you what something unfavorable is:

See how steep that slope is? That's really unfavorable. And that's a factor of five! We don't want the q value to decay too rapidly; I think this would lead to unfavorable results. While I'm not going to post it, I'm sure you can imagine how horrible the "factor of 10" decay would look... in short, very steep!

So why do I care about q-decay, you might ask? Well, currently (graphically speaking) there is a vertical bound on the size of clusters. I am going to implement a horizontal bound as well. My decay will help me enforce that bound. I want the decay to be gentle, so that cluster sizes are not too disparate. In a few weeks, you'll see what I mean, but I'm being a bit mum on this :-) (And I'm having a lot of fun!)

That's all for now. I might upload some diagrams to the webspace sometime soon. I wil let you know then so you can go and see my progress ^-^