# 📦 `{fdp}` R package

[![Project
Status](https://www.repostatus.org/badges/latest/active.svg)](https://www.repostatus.org/#active)
[![License](https://img.shields.io/badge/license-GPL%20%28%3E%3D%203%29-brightgreen.svg?style=flat)](https://www.gnu.org/licenses/gpl-2.0.html)
[![CRAN
version](https://www.r-pkg.org/badges/version/fdp)](https://cran.r-project.org/package=fdp)
[![CRAN
downloads](https://cranlogs.r-pkg.org/badges/grand-total/fdp)](https://cran.r-project.org/package=fdp)
[![R-universe status
badge](https://louisaslett.r-universe.dev/badges/fdp)](https://louisaslett.r-universe.dev/fdp)

The [fdp](https://fdp.louisaslett.com/) package provides tools for
working with f-differential privacy, a powerful framework that
generalises traditional differential privacy definitions. The package
allows you to construct, visualise, and analyse trade-off functions that
show the optimal relationship between Type I and Type II errors when
distinguishing between outputs from neighboring datasets. This is an
initial release with basic functionality that may be extended over time
with composition results and privacy accountants. At present it
hopefully helps researchers in the area quickly produce the f-DP
visualisations used in papers etc. Please cite this software if you use
it.

The package is based on the f-differential privacy/Gaussian differential
privacy framework introduced by Dong, Roth, and Su (2022) in “Gaussian
Differential Privacy” (*Journal of the Royal Statistical Society Series
B*, <https://doi.org/10.1111/rssb.12454>). A nice overview of Gaussian
differential privacy is available in Gomez *et al.* (2025)
<https://arxiv.org/abs/2503.10945>.

## Installation

You can install the package from CRAN:

``` r
install.packages("fdp")
```

Or install the development version from GitHub:

``` r
# install.packages("remotes")
remotes::install_github("louisaslett/fdp")
```

## Example

Here are some basic examples of using the
[fdp](https://fdp.louisaslett.com/) package:

### Plotting Gaussian Differential Privacy

``` r
library(fdp)

# Plot a single GDP trade-off function with μ = 1
fdp(gdp(1))
```

### Comparing Privacy Mechanisms

``` r
# Compare Gaussian DP with classical (ε, δ)-DP
fdp(
  "Gaussian DP" = gdp(1.0),
  "Classical DP" = epsdelta(1.0),
  "Approximate DP" = epsdelta(1.3, 0.05),
  .legend = "Privacy Mechanism"
)
```

### Estimating Privacy Parameters

``` r
# Define empirical trade-off points from a privacy audit
empirical_points <- data.frame(
  alpha = c(0.00, 0.05, 0.10, 0.25, 0.50, 1.00),
  beta  = c(1.00, 0.93, 0.87, 0.72, 0.43, 0.00)
)

# Find the GDP parameters that lower bound these points
gdp_bound <- est_gdp(empirical_points)
gdp_bound

# Visualize the fit
fdp(empirical_points, gdp_bound)
```

### Working with the Laplace Mechanism

``` r
# Create trade-off function for Laplace mechanism with scale parameter 1.5
lap_mechanism <- lap(1.5)

# Find its GDP lower bound
gdp_equiv <- est_gdp(lap_mechanism)

# Compare them
fdp(lap_mechanism, gdp_equiv)
```

For more detailed examples and documentation, see the package vignettes
and function help pages.

# Package index

## All functions

- [`epsdelta()`](https://fdp.louisaslett.com/reference/epsdelta.md) :
  (epsilon, delta)-differential privacy trade-off function
- [`est_epsdelta()`](https://fdp.louisaslett.com/reference/est_epsdelta.md)
  : (epsilon, delta)-differential privacy parameters lower bounding
  empirical trade-off points
- [`est_gdp()`](https://fdp.louisaslett.com/reference/est_gdp.md) :
  Gaussian differential privacy parameters lower bounding empirical
  trade-off points
- [`fdp()`](https://fdp.louisaslett.com/reference/fdp.md) : Plot
  f-differential privacy trade-off functions
- [`fdp_attributes()`](https://fdp.louisaslett.com/reference/fdp_attributes.md)
  [`fdp_line()`](https://fdp.louisaslett.com/reference/fdp_attributes.md)
  [`fdp_point()`](https://fdp.louisaslett.com/reference/fdp_attributes.md)
  [`fdp_name()`](https://fdp.louisaslett.com/reference/fdp_attributes.md)
  : Control rendering of f-DP trade-off functions
- [`gdp()`](https://fdp.louisaslett.com/reference/gdp.md) : Gaussian
  differential privacy trade-off function
- [`gdp_to_epsdelta()`](https://fdp.louisaslett.com/reference/gdp_to_epsdelta.md)
  : Convert Gaussian differential privacy to classical (epsilon,
  delta)-differential privacy
- [`lap()`](https://fdp.louisaslett.com/reference/lap.md) : Laplace
  differential privacy trade-off function
- [`` `+`( ``*`<fdp_plot>`*`)`](https://fdp.louisaslett.com/reference/plus-.fdp_plot.md)
  : Combine fdp plots