9  Supervised Learning

  .:rtemis 0.99.98 🌊 aarch64-apple-darwin20

All supervised learning in rtemis is done using the train() function.
train() can perform:

• Regression & Classification tasks are detected automatically, depending on the outcome variable’s type (numeric => Regression, factor => Classification)
• Automatic hyperparameter tuning: If more than one value is passed for a tunable hyperparameter, train() will automatically perform hyperparameter tuning, a.k.a. model selection.
• Training-test set splitting: Argument outer_resampling defines the resampling used for assessmenet of model performance.

9.1 Overview

Figure 9.1: Overview of the train() function

9.2 Conventions

• The input dataset columns should be either numeric or factors.
• The last column of the input data is the outcome variable.
• If the outcome variable is numeric, regression is performed.
• If the outcome variable is a factor, classification is performed.
• In binary classification, the second factor level is considered the positive case.

9.3 Available Algorithms

Print available algorithms:

available_supervised()

          CART: Classification and Regression Trees
           GAM: Generalized Additive Model
           GLM: Generalized Linear Model
        GLMNET: Elastic Net
      Isotonic: Isotonic Regression
     LightCART: Decision Tree
      LightGBM: Gradient Boosting
       LightRF: LightGBM Random Forest
  LightRuleFit: LightGBM RuleFit
        Ranger: Ranger Random Forest
     LinearSVM: Support Vector Machine with Linear Kernel
     RadialSVM: Support Vector Machine with Radial Kernel
        TabNet: Attentive Interpretable Tabular Learning

Note

The new rtemis focuses on a a tighter integration with a smaller number of algorithms, at least initially. The original version supported a long list of algorithms that we ourselves never ended up using. Support for more algorithms, including our own, is planned and will be continuously reassessed.