differential-privacy-clj

A Clojure wrapper for the Java differential-privacy library.

See online demo notebooks.

Installation

Steps to build it locally:

1. Make sure you have Maven, Bazel and Leiningen installed.

2. Download and build the Java differential-privacy library:

git clone https://github.com/google/differential-privacy.git
cd differential-privacy/java/
bazel build "..."

3. Install libdifferentialprivacy.jar and libsummary-proto-speed.jar in your local Maven repository:

mvn install:install-file -Dfile=bazel-bin/main/com/google/privacy/differentialprivacy/libdifferentialprivacy.jar -DgroupId=com.google.privacy.differentialprivacy -DartifactId=libdifferentialprivacy -Dversion=1.0 -Dpackaging=jar

mvn install:install-file -Dfile=bazel-bin/external/com_google_differential_privacy/proto/libsummary-proto-speed.jar -DgroupId=com.google.differentialprivacy -DartifactId=libsummary-proto-speed -Dversion=1.0 -Dpackaging=jar

4. Clone this repository and install the library:

git clone https://github.com/OpenMined/clojure-dp.git
cd clojure-dp
lein install

Usage

See demo notebooks (run online).

Examples:

(require '[differential-privacy-clj.core :as dp])

(def example-data (take 1000 (repeatedly #(rand 10.0))))

(println "True sum: "
         (reduce + example-data))
(println "DP sum:   "
         (dp/bounded-sum example-data :epsilon 0.1
                         :lower 0 :upper 10
                         :max-partitions-contributed 1))

(println "True count:"
         (count example-data))
(println "DP count:  "
         (dp/count example-data :epsilon 0.1
                   :max-partitions-contributed 1))

(println "True mean:" (/ (reduce + example-data)
                         (count example-data)))
(println "DP mean:" (dp/bounded-mean example-data :epsilon 0.1
                                     :lower 0 :upper 10
                                     :max-partitions-contributed 1
                                     :max-contributions-per-partition 1))

will print something like:

True sum:  4988.542973798648
DP sum:    5175.075793958153
True count: 1000
DP count:   999
True mean: 4.988542973798648
DP mean: 5.002603661455349

Available algorithms:

✔️ - Supported ✅ - Planned

Algorithm
Count	✔️
Sum	✔️
Mean	✔️
Variance	✅
Standard deviation	✅
Order statistics (incl. min, max, and median)	✅

Support

For support in using this library, please join the #lib_dp_java Slack channel. If you’d like to follow along with any code changes to the library, please join the #code_dp_clojure Slack channel. Click here to join our Slack community!