Okay. We're going to look at random forest now, and how to use random forest from H2O. I'm using the same setup as we looked at last week for deep learning. So, is the Iris data set, and I've run these commands, on the same over in Python. So, we have 118 training rows, 32 testers. Now to do deep learning, we did this command, didn't we? From h2o.estimators.deeplearning, import H2ODeepLearningEstimator, create an instance of it called Co train. And for random forest, let's just get past these. For random forest it's h2oestimators.random_forest, Import H2ORandomForestEstimator. Let's go. Exactly the same command. Train, first give the columns we want to use as inputs, then the column we want as the correct answer, the thing we're learning, and our training data. Finished while I was still talking about it, let's see what we've got. Seven errors on the training data, which is 94 percent. Okay. This is how many trees it's used. So it's made a forest of 50 trees in 0.23 seconds. And you can see this is the error. So the error started off as 0.4 with just one tree, point three, after a second tree, coming down to 0.19. Predict on your random forest model and model performance, let's see how it does on data it hasn't seen. And it has done well, only one mistake out of 32, which is 96.8 percent. Let us scroll back up 94 on the data it saw, 96 on the data hadn't. So, no over learning generalized very well. Let's just go look at the commands in R. So this was what we used for deep learning. To use random forest, we change the web deep learning to random forest. And when you are using this, we have type ahead, so there we go, RA was enough to find the correct command. Where we go and run it earlier we used 50 trees, it got six errors on 118 which was 95 percent, correct. And again, it got one wrong out of 32, 97 percent on the unseen test data, so very good results. And random forest is a good choice of algorithm for a categorical data. When you want to learn about all the options in how you do question mark and then the name of the function, H2O.randomforest. And as we're using our studio, instead of showing in the column so, it shows in the help tab. Nicely formatted all the various parameters. So far we've only been setting X Y and training frame. You can see there's a lot here. And they're all explained in detail here, or explained in the manuals that you'll find on the H2O website. The Python equivalent of that is a bit longer, help and then open brackets and then you need to get the full path to the name of the estimator.
