1 item has been added to your cart.

Learn about H2O machine learning in Stata.

Ensemble decision trees: Gradient boosting machine (GBM) and random forest

• GBM for regression for continuous and count responses
• GBM for binary classification
• GBM for multiclass classification
• Random forest for regression
• Random forest for binary classification
• Random forest for multiclass classification
• Many loss functions for GBM models
• Many encoding schemes for categorical variables
• Monotonicity constraints on predictors in GBM models
• Model selection using cross-validation
• Early stopping

Hyperparameter tuning

• Select best-performing model by tuning
• Number of trees
• Learning rate of each tree in GBM models
• Learning rate decay in GBM models
• Maximum depth of each tree
• Minimum number of observations for splitting a leaf node
• Sampling rate for selecting predictor subset per tree in GBM
• Sampling value for selecting the number of predictors in random forest
• Sampling rate for selecting observations per tree
• Minimum node-split threshold
• Number of histogram bins for continuous and categorical predictors
• Many tuning metrics for regression and classification analysis
• Two grid-search methods: Cartesian and random
• Different early-stopping methods for random grid search

Tuning and estimation summaries

• Display various model performance metrics
• Summarize cross-validation results
• Summarize results from hyperparameter grid search
• Select the best model after performing a grid search
• Explore alternative models after grid search
• Compare goodness of fit for machine learning models
• Plot score history

Model performance evaluation

• Binary classification
• Display a confusion matrix
• Display threshold-based metrics
• Produce receiver operating characteristic (ROC) curve plot
• Produce precision–recall curve plot
• Multiclass classification
• Display a confusion matrix
• Display area under the curve (AUC) and area under the precision–recall curve (AUCPR)
• Display hit-ratio tables

Postestimation frame and estimation results

• Define frame for postestimation analysis
• Store and restore model estimation results