Custom Synthesis Method

To create your own synthesizer, you need to implement a Python class with three required methods and configure the YAML file to use it.

Usage Examples

Click the below button to run this example in Colab:

Loader:
  load_benchmark_with_schema:
    filepath: benchmark://adult-income
    schema: benchmark://adult-income_schema
Synthesizer:
  your-custom-method:
    method: custom_method
    module_path: custom-synthesis.py  # Python file name
    class_name: MySynthesizer_Shuffle # Class name in the file

Required Implementation

Your Python class must have:

class YourSynthesizer:
    def __init__(self, config: dict, metadata):
        """Initialize your synthesizer"""
        pass

    def fit(self, data: pd.DataFrame):
        """Learn from the input data"""
        pass

    def sample(self) -> pd.DataFrame:
        """Generate and return synthetic data"""
        pass

Example: Shuffle Synthesizer

Our example custom-synthesis.py implements a simple synthesizer that:

1. Stores each column’s values during fit()
2. Shuffles each column independently to break correlations
3. Returns the shuffled data in sample()

This preserves the distribution of each column while removing relationships between columns - useful for simple anonymization or as a baseline.