Exercise: Ensemble Models
The dataset we use in this exercise describes the sale of individual residential properties in Ames, Iowa from 2006 to 2010. One of the columns is the overall condition ranking, with values between 1 and 10. The goal of this exercise is to train a binary classification model, which can predict whether the overall condition is high or low. To do so, the workflow below reads the data set and creates the class column based on overall condition ranking, which is called rank and has the values low if the overall condition is smaller or equal to 5, otherwise high.
It is now on you continue this workflow!
Train a random forest model (Random Forest Learner) to predict the overall condition of a house (high/low), with the following settings:
Examine the output table “Attribute Statistics” generated by the learner node. It lists which features were chosen for the first 3 splits in different tree models. Features chosen first are likely informative in classification. Which of the features in this dataset are very informative? Justify your answer.
Use the trained model to predict (Random Forest Predictor) the rank of the houses in the test set, and evaluate the accuracy of the decision tree model (Scorer JavaScript). What is the accuracy of the model?
Train a gradient boosted trees model to predict the overall condition of a house (high/low), with the following setting:
Use the trained model to predict the rank of the houses in the test set, and evaluate the accuracy of the decision tree model. What is the accuracy of the model?
(Optional) Use a parameter optimization loop to find the optimum tree depth to maximize the accuracy (use Scorer node, not Scorer JavaScript). Examine the tree depth of 1 to 10, with the step size of 1. Which tree depth produces the optimal result?
