Formal ML is a LEAN library for proving foundational results in measure theory, probability, statistics, and machine learning, based upon mathlib. It is in early development, and not all proofs are complete.
It introduces probability spaces, the Radon-Nikodym derivative, PAC bounds, etc.
The library can be used in other packages as well.
This package is compatible with LEAN 3.20.0.
This package is based upon the LEAN community version of lean, specifically version 3.20.0. Before building the package, one must download LEAN 3.20.0.
For Linux versions, you can download https://github.com/leanprover-community/lean/releases/download/v3.20.0/lean-3.20.0-linux.tar.gz. After unzipping the file, you can include the bin subdirectory in your PATH (no need to recompile). To test your framework, run:
lean --version
leanpkg help
To download this package, run:
git clone https://github.com/google/formal-ml.git
cd formal-ml
To build the package, run:
leanpkg build
This will download mathlib, and compile both mathlib and this package, so it may take a few minutes.