PETsARD (Privacy Enhancing Technologies Analysis, Research, and Development, /pəˈtɑrd/) is a Python library for facilitating data generation algorithm and their evaluation processes.

The main functionalities include dataset description, various dataset generation algorithms, and the measurements on privacy protection and utility.

PETsARD（隱私強化技術分析、研究與開發）是一套為了促進資料生成演算法及其評估過程而設計的 Python 程式庫。

其主要功能包括描述資料集、執行各種資料集生成算法，以及對隱私保護和效用進行測量。

• Milestone
  • The Milestone document provide detailed information about the latest version of PETsARD.
  • 里程碑文件包含了 PETsARD 最新版本的詳細資訊。
• CHANGELOG.md
  • The Changelog document provide the changelog for understand the evolution of the PETsARD over time.
  • 歷史文件提供了提供了變更日誌，以便了解 PETsARD 隨時間演進的情況。
• PETsARD - User Guide
  • The User Guide aims to assist developers in rapidly acquiring the necessary skills for utlisiing PETsARD in data synthesis, evaluating synthesized data, and enhancing the research efficiency in Privacy Enhancing Technologies-related fields.
  • 使用者指南旨在幫助開發者迅速獲得必要的技能，以使用 PETsARD 進行數據合成、合成數據的評估，以及提升開發者隱私增強相關領域的研究效率。

Installation 安裝

To install this package, please follow these steps to set up your environment using the requirements.txt file:

要安裝本套件，請依照以下步驟使用 requirements.txt 檔案來設定您的環境：

1. Create a virtual environment and activate it:

2. 建立虛擬環境並啟動：

python -m venv venv
source venv/bin/activate  # Linux/Mac
# or 或是
venv\Scripts\activate     # Windows

2. Upgrade pip:

3. 升級 pip：

python -m pip install --upgrade pip

3. Install the required packages using requirements.txt:

4. 使用 requirements.txt 安裝所需套件：

pip install -r requirements.txt

Please note: It is important to use the exact versions specified in our requirements.txt file to ensure compatibility and avoid dependency conflicts. Our team has tested and confirmed these specific versions work well together:

請注意：請務必使用我們在 requirements.txt 中指定的版本，以確保相容性並避免相依性衝突。我們的團隊已測試並確認這些特定版本能夠良好地一起運作：

sdv==1.12.1
smartnoise-synth==1.0.3
anonymeter==1.0.0
sdmetrics==0.14.0
pyyaml==6.0.1
requests
boto3
git+https://github.com/ryan112358/private-pgm.git

Getting Started 入門指南

We recommend that users write the experimental planning in YAML format, which details in the PETsARD - User Guide, and use Executor in PETsARD to conduct the experiment.

我們建議使用者將自己的實驗規劃撰寫成 YAML 格式，呼叫 PETsARD 的 Executor 讀取執行實驗，以上的細節可參見 PETsARD - User Guide。

Executor

Here is the simplest way to get started with PETsARD:

以下是最簡單入門的 PETsARD 使用方式：

from PETsARD import Executor

filename = "Exec_Design.yaml"
exec = Executor(config=filename)
exec.run()

YAML

Here, we use the default methods of each module to construct the simplest 'Exec_Design.yaml'. For specific settings of each module, please refer to the YAML page.

這裡利用各模組的預設方法 (default) 建構最簡單的 'Exec_Design.yaml'，各模組具體設定，請參考 YAML 頁面。

---
Loader:
  demo:
    # default of Loader is Adult Income dataset
    method: 'default'
Preprocessor:
  demo:
    # default of Preprocessor automatically determines the data type
    #   and uses the corresponding method.
    method: 'default'
Synthesizer:
  demo:
    # default of Synthesizer is SDV Gaussian Copula
    method: 'default'
Postprocessor:
  # make sure the expt_name of Postprocessor is same as the Preprocessor
  demo:
    method: 'default'
Evaluator:
  demo:
    # defalut of Evaluator is SDMetrics QualityReport
    method: 'default'
Reporter:
  save_data:
    method: 'save_data'
    output: 'YAML Demo'
    # source of Reporter means which result of module/expt_name should Reporter use
    #   accept string (for only one) and list of string (for multiple result)
    source: 'Postprocessor'
  save_report_global:
    method: 'save_report'
    output: 'YAML Demo'
    # eval in Reporter means which
    #   expt_name of Evaluator/Describer should Reporter use
    eval: 'demo'
    # granularity = 'global' indicates that
    #   the scoring covers the entire dataset with a comprehensive level of detail.
    granularity: 'global'
...

Modules 各模組

• Loader: The module for data loading, see Loader page.
  • Benchmark datasets: The usage of Loader to download the benchmark datasets, see Benchmark datasets page.
• Splitter: The module for splitting a dataset into training and validation datasets, see Splitter page.
• Processor: The module for data pre/postprocessing, see Processor page.
• Synthesizer: The module for data synthesising, see Synthesizer page.
• Evaluator: The module for synthetic data evaluation, see Evaluator page.
• Describer: The module for data description, see Describer page.
• Reporter: The module for data saving and report output, see Reporter page.

• Loader：讀取資料的模組，見 Loader 頁面。
  • Benchmark datasets：如何利用 Loader 獲取預先載入的基準資料集，見 Benchmark datasets 頁面。
• Splitter：把資料切分成實驗組與控制組的模組，見 Splitter 頁面。
• Processor：對資料進行前處理跟後處理的模組，見 Processor 頁面。
• Synthesizer：對資料做合成資料等隱私強化處理的模組，見 Synthesizer 頁面。
• Evaluator：對合成資料結果做評估的模組，見 Evaluator 頁面。
• Describer：對資料本身做描述的模組，見 Describer 頁面。
• Reporter：對資料進行存檔、以及資料評估與描述輸出報告的模組，見 Reporter 頁面。

Contributing 貢獻

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change. And please make sure to update tests as appropriate.

歡迎提交拉取請求。對於重大變更，請先開啟一個議題來討論您想做的改變。並請確保適當地更新測試。

License 授權

This project is licensed under MIT License with additional restrictions due to dependencies. The most significant restriction comes from SDV's Business Source License 1.1, which prohibits using this software as a commercial synthetic data service. For detailed licensing information, please see the LICENSE file.

本專案採用 MIT 授權，但因相依套件而有額外限制。最主要的限制來自 SDV 的 Business Source License 1.1，禁止將本軟體用於商業性的合成資料服務。詳細的授權資訊請參閱 LICENSE 檔案。

Key Dependencies' Licenses 主要相依套件授權：

• SDV: Business Source License 1.1
• Anonymeter: The Clear BSD License
• SDMetrics: MIT License
• Smartnoise: MIT License

For commercial use involving synthetic data services, please contact DataCebo, Inc.

如需將本軟體用於合成資料的商業服務，請聯絡 DataCebo, Inc.

Citation 引用

• Synthesizer module:
  • SDV - sdv-dev/SDV:
    • Patki, N., Wedge, R., & Veeramachaneni, K. (2016). The Synthetic Data Vault. IEEE International Conference on Data Science and Advanced Analytics (DSAA), 399–410. https://doi.org/10.1109/DSAA.2016.49
  • smartnoise - opendp/smartnoise-sdk:
• Evaluator module:
  • Anonymeter - statice/anonymeter:
    • Giomi, M., Boenisch, F., Wehmeyer, C., & Tasnádi, B. (2023). A Unified Framework for Quantifying Privacy Risk in Synthetic Data. Proceedings of Privacy Enhancing Technologies Symposium. https://doi.org/10.56553/popets-2023-0055
  • SDMetrics - sdv-dev/SDMetrics