Skip to content

Quick Start

This path gets from a clean Python environment to a first VertiMosaic run without requiring a GPU or private data.

Requirements

  • Python 3.11 or 3.12
  • a CPU environment
  • pip

Install the released package

python -m venv .venv
. .venv/bin/activate  # Windows: .venv\Scripts\activate
python -m pip install --upgrade pip
python -m pip install vertimosaic

Confirm the CLI is available:

vertimosaic --help

Run the first deterministic demo

vertimosaic demo --rows 2000 --seed 42

The demo is a smoke experiment intended to verify installation and the basic VFL execution path. Research benchmark paths use stronger evidence generation and larger bootstrap settings where practical.

Run the formal comparison

vertimosaic-benchmark-matrix

This is the main centralized / single-party / VFL comparison entry point. Use the generated artifacts rather than manually transcribing metrics when reporting results.

Create an independent reproduction bundle

For a released version such as 0.3.0:

python -m pip install vertimosaic==0.3.0
vertimosaic-reproduce --expected-version 0.3.0 --output evidence

The evidence directory is designed for inspection and independent sharing. See Independent Reproduction for the expected artifacts and attestation model.

Optional privacy research dependencies

PSI and Paillier adapters are intentionally optional:

python -m pip install "vertimosaic[privacy-crypto]"

Installing these dependencies does not make every VertiMosaic protocol cryptographically private. Read Privacy Boundaries, Threat Model, and Research Backends before describing the guarantee of an experiment.

Work from the repository

git clone https://github.com/sauravsingla/VertiMosaic.git
cd VertiMosaic
python -m pip install --upgrade pip
python -m pip install -e ".[dev]"

Run the verification suite:

make coverage
make coverage-protocol

The CI also checks formatting, linting, static typing, package build integrity, protocol-critical coverage, reproducibility smoke paths, privacy audits, and serialized transport behavior.

Choose the next path

Goal Next step
Understand party/model boundaries Architecture
Understand message flow and alignment Protocol
Evaluate privacy guarantees Privacy Boundaries
Select an appropriate dataset Benchmarks
Reproduce a release Reproducibility
Use public classes directly API Reference

Focused tutorials

The repository also contains four hands-on tutorials:

  1. First VFL in 10 minutes
  2. Two-party logistic VFL
  3. Histogram GBDT
  4. Privacy leakage and mitigation

For a separate-service developer-laptop example, see the Docker Compose HTTPS/mTLS deployment.