contributing.md

Contributing to SmartNoise

Contributions to SmartNoise are welcome from all members of the community. This document is here to simplify the on-boarding experience for contributors, contributions to this document are also welcome. Please use the GitHub issue board to track and take ownership of issues.

Please let us know if you encounter a bug by creating an issue. We appreciate all contributions. We welcome pull requests with bug-fixes without prior discussion.

If you plan to contribute new features, utility functions or extensions to the core, please first open an issue and discuss the feature with us. Sending a PR without discussion might end up resulting in a rejected PR, because we might be taking the core in a different direction than you might be aware of.

For a description of the library architecture and installation instructions, see smartnoise-core/README.md. Before contributing, we recommend following the Getting Started guide in that README to get sample notebooks running and to gain some familiarity with the library structure.

General Contribution Guidelines

• Use GitHub issues including the Github issue board.
• Anything implemented from the differential privacy literature should contain clear citations and/or whitepapers explaining any necessary derivations.
• Code should be clearly documented and include testing. For more information, please see:
  • Block tests in Rust: https://doc.rust-lang.org/rustdoc/documentation-tests.html
  • Testing modules in Rust: https://doc.rust-lang.org/book/ch11-03-test-organization.html

Contributing to the Validator

Components need to be added to both the protobuf and to the source codebase.

1. Add a component to protobuf in validator-rust/prototypes:

Add a JSON file to the components subdirectory. An entry in components.proto will be made upon build.

2. Add component to src/components:

This will look different depending on the type of component you are contributing. The following list describes traits that must be implemented for common types of components.

For new statistics:

• Include a Component implementation containing the propagate_property function which how to propagate properties through that node in the graph. This function also checks that differential privacy guarantees are met: e.g. that the sensitivity is computable, that the data is conformable, and/or overflows can cause the computation to fail.
• A Sensitivity implementation with a compute_sensitivity function that describes how to compute the statistic for all combinations of implemented privacy definitions and sensitivity spaces. - Any derivations used for sensitivities should be derived in the whitepapers/sensitivities, or there should be a citation to published sensitivity analysis (or both). - Include a comment in the Sensitivity implementation that links to the location of the proof

For new DP statistics:

• An Expandable implementation with an expand_component function which describes how to expand the computation graph to insert each component of the differentially private calculation.
• A Report implementation with a summarize function that stores the results of the differentially private computation as JSON.

#For new mechanisms:

• A Component implementation containing a propagate_property function describing how to propagate input properties through that node in the graph. This also contains checks on the privacy parameters to verify that they are reasonable.
• An Expandable implementation with an expand_component function which describes how to expand the computation graph
• If possible, an Accuracy implementation with accuracy_to_privacy_usage and privacy_usage_to_accuracy functions that describe how to transition between accuracy and privacy guarantees. Any associated derivations should be recorded in whitepapers/accuracy.
• Contributors should add a whitepaper on the mechanism itself in whitepapers/mechanisms or clearly cite the academic paper the mechanism originated from.

Contributing to the Rust Runtime

Each Component in the Rust runtime (runtime-rust) requires an Evaluable trait that describes how to evaluate a node of the computation graph. Any new component should include documentation for both the arguments to the evaluate function and the the return type(s). Additionally, doctests and test modules should be included.

In order to contribute to underlying mechanisms and noise selection, which are implemented in the crate's utility functions, first read the noise whitepaper in whitepaper/noise to understand the current design choices. Clearly document and test any code.

Documentation and Testing

Code contributions should include both doctests and unit tests. We also encourage contributors to run the samples in smartnoise-samples and to write their own samples that highlight their contributions. All code is integration tested and reviewed before merging.