 In our last video, we talked about classification trees. I used the datasets widget to load the familiar iris dataset. Then I constructed a tree and examined it with the tree viewer. And that looked more or less like this. Now, a classification tree is grown from the root node, here at the top. It includes the entirety of the iris trading set. Then, from there, the tree growing algorithm finds the best feature to split the data into two separate groups with distinct classes. Now, the first split involves petal length. If it's shorter than 1.9 centimeters, the irises are setosas. And on the other hand, virginica and versicolors have longer petal lengths. So, all of those wind up in the right node. Now, we want to continue to split this subset of 100 irises even further. So, in the next step, we'll compare petal widths, as it seems that versicolors tend to have somewhat narrower petals than virginicas. So, here, the algorithm decided to not split the node on the right any further. Apparently, it's okay, with a single versicolor clump together with 45 virginicas. However, the left node contains 49 versicolors and 5 virginicas. So, it does continue to split them further. And again, it compares petal lengths and sets 4.9 centimeters as the threshold. So, how does orange know when to stop growing a tree? Well, in the case of the setosas, it's pretty clear, since there are no other classes in the node. But, in every other situation, there are rules. So, let's open up the tree widget. Here, you can see and adjust just how orange grows the trees. By default, it infers binary trees. It stops growing the tree if the leaves contain less than 2 instances. And it doesn't split nodes with less than 5 instances. Now, these parameters define when to stop growing a tree. This is known as forward pruning. So, now I can change some of these settings and see how it affects the resulting tree structure. First, I'll place the tree and tree viewer widget side by side to get a better look. Now, let's see what happens when I change the minimum number of instances in leaves to 4. The tree just got smaller. It now only contains 3 internal nodes and 4 leaves. Now, I can also make the tree stop growing at a depth of 2. This way, we get the simplest tree that separates the 3 classes pretty well. So, I'm just going to reset these two parameters to their default values of 2 and 100 to get the widget ready for my next video. Just like all other widgets in orange, the tree also remembers what settings I used. And it's worth checking out the parameters values every time I do some modeling. Because the predictive behavior of my model almost always depends on these parameters. And soon, we'll talk some more about that. But let's take it slow. We'll continue in our upcoming video.