3.15.11 Model Types

HeatWave AutoML supports the following training models. When training a model, use the ML_TRAIN model_list and exclude_model_list options to specify the training models to consider or exclude. Section 3.15.12, “Model Metadata” includes the algorithm_name field which defines the model type.

• Classification models:

  • LogisticRegression.

  • GaussianNB.

  • DecisionTreeClassifier.

  • RandomForestClassifier.

  • XGBClassifier.

  • LGBMClassifier.

  • SVC.

  • LinearSVC.

  • ExtraTreesClassifier. Supported as of MySQL 8.1.0.

• Regression models:

  • DecisionTreeRegressor.

  • RandomForestRegressor.

  • LinearRegression.

  • LGBMRegressor.

  • XGBRegressor.

  • SVR.

  • LinearSVR.

  • ExtraTreesRegressor. Supported as of MySQL 8.1.0.

• Forecasting models:

  • Univariate endogenous models:

    • NaiveForecaster.

    • ThetaForecaster. Supported as of MySQL 8.0.32.

    • ExpSmoothForecaster.

    • ETSForecaster. Supported as of MySQL 8.0.32.

    • STLwESForecaster: STLForecast with ExponentialSmoothing substructure. Supported as of MySQL 8.0.32.

    • STLwARIMAForecaster: STLForecast with ARIMA substructure. Supported as of MySQL 8.0.32.

  • Univariate endogenous with exogenous models:

    • SARIMAXForecaster. Supported as of MySQL 8.0.32.

  • Multivariate endogenous with exogenous models:

    • VARMAXForecaster. Supported as of MySQL 8.0.32.

  • Univariate or multivariate endogenous with exogenous models:

    • DynFactorForecaster. Supported as of MySQL 8.0.32.

• Anomaly detection models:

  • The only model is GkNN. See: Section 3.10, “Anomaly Detection”.

• Recommendation models:

  • Recommendation models that rate users or items to use with explicit feedback.

    • Baseline.

    • CoClustering.

    • NormalPredictor.

    • SlopeOne.

    • Matrix factorization models:

      • SVD.

      • SVDpp.

      • NMF.

  • Recommendation models that rank users or items to use with implicit feedback.

    • BPR: Bayesian Personalized Ranking from Implicit Feedback.