Key features of teradataml AutoML:
- Supports different problem types:
- Regression
- Binary Classification
- Multiclass Classification
- Clustering
- Provides the following models for training based on problem types:
- GLM
- SVM
- Decision Forest
- XGBoost
- KNN
- KMeans
- GaussianMixture
- Gives flexibility to select specific models out of available models.
- Performs all five phases in automated way, but can also be customized based on user input.
- Generates model leaderboard and leader for given dataset.
- Provides hyperparameters of trained models.
- Provides prediction on validation dataset as well as user passed data using model leader or any other model from leaderboard.
- Provides early stopping criteria to stop AutoML training before completion time using three different options by defining early stopping timer, early stopping metric threshold, and maximum models to be trained.The number of models generated will be determined by the following:
- Default condition: The default number of models generated by AutoML if no early stopping criteria are used.
- Early stopping conditions: When early stopping conditions are utilized by the user, the final count of models will depend on which of these conditions is satisfied first:
- The number of models trained within the specified early stopping timer.
- The number of models trained to achieve the specified early stopping performance metric threshold.
- The number of models specified under the maximum models to be trained.
- Provides three different logging levels to display required contents, higher level provides more detailed output.
- Provides three feature selection methods as part of data preparation for non-clustering problem: RFE, Lasso, and PCA. By default, RFE and PCA are executed, while Lasso can be optionally enabled through a user-defined flag enable_lasso for performing Lasso-based feature selection.
- Generates a default ID column named automl_id if you don't specify one. Otherwise, it uses the user-provided ID column consistently throughout the training and prediction mapping processes.
- Provides the preserve_columns flag to retain specific columns from the transformed data in the prediction dataframe. If not set, only the ID and target columns are preserved by default for prediction.
- Provides the raise_errors flag to control error-handling behavior; when set to True, non-blocking errors are raised as errors, otherwise they are logged as warnings.
AutoML progress bar uses Unicode characters and may return an error in non-UTF environments (e.g., Windows cp1252). Use a UTF-8 enabled platform to avoid this issue.