Docs
Getting Started
Data Synthesis with Default Parameters

Data Synthesis with Default Parameters

The simplest way to generate privacy-enhanced synthetic data. Current default synthesis uses built-in Gaussian Copula implementation.

Click the below button to run this example in Colab:

Open In Colab

Note: If using Colab, please see the runtime setup guide.

Loader:
  load_csv:
    filepath: benchmark/adult-income.csv
Preprocessor:
  default:
    method: 'default'
Synthesizer:
  default:
    method: 'default'
Postprocessor:
  default:
    method: 'default'
Reporter:
  output:
    method: 'save_data'
    source: 'Synthesizer'

YAML Parameters Detailed Explanation

Loader (Data Loading Module)

  • load_csv: Experiment name, can be freely named, recommended to use descriptive names
  • filepath: Data file path
    • Value: benchmark/adult-income.csv
    • Description: Specifies the location of the data file to load. This example uses adult-income.csv, which you can replace with your own CSV file path
    • Supported formats: CSV, TSV, Excel (requires openpyxl), OpenDocument
    • Supports relative or absolute paths
    • Also supports benchmark:// protocol for automatic standard dataset downloads

Recommended: Use Schema: To ensure data loading accuracy and consistency, it is highly recommended to use the schema parameter to pre-define the data structure. Schema allows you to explicitly specify the data type of each column (numeric, categorical, datetime, etc.), constraints, and relationships between columns.

Example with schema:

Loader:
  load_csv:
    filepath: benchmark/adult-income.csv
    schema: benchmark/adult-income_schema.yaml

For detailed information about Schema, please refer to the Schema YAML Documentation.

Preprocessor (Data Preprocessing Module)

  • default: Experiment name, can be freely named
  • method: Preprocessing method
    • Value: default
    • Description: Uses the default processing sequence, including the following steps:
      1. missing (missing value handling): Numeric columns use mean imputation, categorical columns are dropped
      2. outlier (outlier handling): Numeric columns use IQR method
      3. encoder (encoding): Categorical columns use uniform encoding (encoder_uniform)
      4. scaler (scaling): Numeric columns use standardization (scaler_standard)

Synthesizer (Synthetic Data Generation Module)

  • default: Experiment name, can be freely named
  • method: Synthesis method
    • Value: default
    • Description: Uses the default synthesis method, which is PETsARD Gaussian Copula
    • Gaussian Copula is a statistical-based synthesis method that captures correlations between variables
    • This is a built-in implementation, no external dependencies required

Postprocessor (Data Postprocessing Module)

  • default: Experiment name, can be freely named
  • method: Postprocessing method
    • Value: default
    • Description: Automatically performs reverse operations of Preprocessor to restore synthetic data to original format
    • Restoration sequence (reverse of preprocessing):
      1. inverse scaler: Reverse scaling, restoring standardized values
      2. inverse encoder: Reverse encoding, restoring encoded categorical variables
      3. restore missing: Reinsert missing values according to original proportion
    • Note: Outlier handling cannot be reversed

Reporter (Output Module)

  • output: Experiment name, can be freely named
  • method: Report method
    • Value: save_data
    • Description: Saves the output data from the specified module as a CSV file
  • source: Data source module
    • Value: Synthesizer
    • Description: Saves synthetic data generated from the Synthesizer module
    • Can also choose other modules, such as Preprocessor, Postprocessor, etc.
    • Default output file naming format: petsard_Synthesizer[output].csv

Execution Flow

  1. Loader loads the adult-income.csv data
  2. Preprocessor performs preprocessing (impute missing values, handle outliers, encode, scale)
  3. Synthesizer generates synthetic data using the Gaussian Copula method
  4. Postprocessor restores synthetic data to original format (reverse scaling, reverse encoding, insert missing values)
  5. Reporter saves the final synthetic data as a CSV file

Advanced Usage

For custom parameters, please refer to the following documentation:

Getting StartedData Synthesis and Evaluation with Default Parameters