HyperModel is an abstract class that needs to implement a dedicated HyperModel for different frameworks or domains. HyperModel explore hyper-parameters sample from Searcher, fit and evaluate Estimator, then reward the metric score to Searcher for optimization. The figure below shows HyperModel search sequence.
To customize HyerModel, two components are required:
- HyperModel: subclass of hypernets.model.HyperModel, create newer Estimator instance with searched space sample, and load trained estimator from storage.
- Estimator: subclass of hypernets.model.Estimator, the core component for model fitting/evaluation/prediction/persistence.
You can reference hypernets.examples.plain_model.PlainModel and hypernets.examples.plain_model.PlainEstimator as start point. See DeepTables, HyperGBM, HyperKeras for more details.