experiment-data-builder

Generating data loaders from an experiment plan involves converting plan specifications into executable data generators and loaders. This process outputs @register_dataset classes and Hydra YAML configs for your pipeline.

Creating Hydra configs for reproducible ML datasets is done by translating experiment-plan specifications into Hydra-ready dataset definitions. This ensures generated dataset classes and validation scaffolds support end-to-end reproducible pipelines.

Yes, you can bootstrap synthetic dataset loaders for ablation studies by using a data builder to translate experiment plans. It generates the necessary dataset classes and validation scaffolds to support reproducible ML experiments.

Yes, the data generation approach works with Factory or Registry patterns by generating @register_dataset classes. These classes register automatically with the Factory or Registry from your project-scaffold to ensure seamless integration.

To start building data generators from plan specs, you need an experiment-plan specification detailing Datasets and Splits. Running the data builder on this plan generates the required dataset classes, Hydra configs, and validation scaffolds.