Before we conclude, I want to give you an overview of all the methods that we covered in this week. As you may recall, what we have done here is basically to talk about networks first, and then followed by trees, which is a particular kind of network. Now, within networks, general networks, we saw that there are two main classes of methods. The first one is node-link diagrams, and the second one is the matrix visualization. Now, we can also say that there is a parallel with trees. It is kind of like the same with trees. With trees, you can visualize trees using node-link diagrams, a different type of node-link diagrams, but node-link diagrams nonetheless, and the second class is containment and partitioning. Now, going back to general networks, we saw that within the class of node-link diagrams, we split them into additional classes. The first one is a force-directed layout where there is an algorithm that positions the nodes in a way to make the structure, the topology of the network, visible. So, that's the main goal of these methods, whereas, the second class of methods is all those network visualization methods, where the nodes are constrained and are positioned in a layout that is fixed or somewhat fixed. Now, we also saw that within trees in the class of containment and partitioning, we have a number of alternative techniques. In particular, I introduced tree maps, which are by far the most popular technique within this class, and I've also introduced some less popular but interesting techniques, which are the sunburst visualization and the icicle visualization. Now, before concluding, I want to come up with a number of observations, or let say, reflection points, related to all the methods that we have seen, their advantages and disadvantages. So, point one, I would say that node-link diagrams are very good for visualizing structure, but often they create clutter, especially for general networks. So, it's important to keep in mind this idea, that if your goal is to visualize the structure, node-link diagrams are probably most of the time, the best solution. We also saw that when node-link diagrams are used for visualizing trees, they are also very good for communicating information about the structure, and they are also very intuitive. That's another very good positive aspect of node-link diagrams. People naturally perceive and understand what the visualization is showing. The downside, is that they do not scale well. The more nodes and links you have, the harder it is to put them all together within the space constraints that you have. So, one thing to keep in mind for node-link diagrams, is that the nodes and the links that are drawn can be used to encode additional information that is related to the objects or the elements that are associated to the nodes and the links. In particular, what happens is that, links can be, let's say, enhanced with visual encoding methods such as length, width, or particular patterns to associate to the line. The nodes can encode information using typically size, color, and shape. So, whenever this is appropriate, I think it's good practice to think about what kind of information associated to the nodes and the edges, can be visualized directly using these encoding strategies. Now, going to the matrix visualization, what we saw is that one of the biggest advantages of matrices, is that they have very good visibility properties. Why? Well, simply because you don't have links and nodes, you don't have line crossings. So, it's much harder to create a sense of clutter. It's much much higher visibility. One downside of matrices, however, is that in order to see any interesting patterns, you have to be mindful of the way the rows and the columns are reordered. So, typically, they need some kind of algorithm or strategy to be reordered in a meaningful way. Another problem of matrices is, that as the number of nodes grows, they tend to have cells that get smaller and smaller and smaller and smaller. So, you can't really afford visualizing too many nodes at once. Now, going to fixed layouts, one problem of fixed layout is that they do not visualize the structure of the network as well as force-directed layouts. And some designs need some reordering. So, for instance, the circular layout may show some patterns better or worse according to how the nodes are reordered around the circle. So, that's an issue there. In any case, they tend to have much higher visibility, especially of the nodes, and they are also very expressive because the position of the nodes tend to be very meaningful. Especially, a very special case is when the nodes are positioned according to geographical layout. Now, the position of the node has a very important meaning. Now, going to the case of trees, we saw tree maps. So, tree maps scale very well, much better than the other techniques, but they also have the problem that structure is not readily visible. So, that's the biggest downside of tree maps. We also saw the sunbursts and icicle plots. I would say that these two visual representations still scale well, pretty well, but they also visualize the structure in a much more intuitive and visible way than tree maps. So, I want to conclude by, I want to single out important properties to keep in mind when you are making choices about what is the best representation for the problem that you have in visualizing network or tree data. So, I would say, one big problem to keep in mind is clutter. So, the more nodes and edges you have, the higher the amount of clutter. Related to that, scalability, does your visualization scale to the number of nodes and number of edges that you have? If not, you have to find out to either reduce the number of nodes and edges, or find a way to filter them or find a way to group them together. Another important property is whether visualizing structure is important. As we have seen typically, if structure is important, node-link diagrams tend to fair better than other solutions. We also saw that reordering in some visualizations is very important. Finally, I want to conclude by keeping in mind that there is another problem, the problem of familiarity and intuitiveness. Many of these visualizations, people are very familiar with some of these visual representations, and they are also very intuitive. So, you always have to strike a balance between effectiveness of these visual representations and also how intuitive and familiar they are to the audience and people who are supposed to use it.
