[−][src]Crate ffsvm

FFSVM stands for "Really Fast Support Vector Machine", a libSVM compatible classifier. It allows you to load models trained by libSVM's svm-train, and use them from your Rust code.

Background

Support Vector Machines (SVMs) are a class of relatively simple and fast machine learning algorithms. They have

few parameters (making them easy to tune),
good generalization properties (making them good learners with limited data) and
overall good classification accuracy.

LibSVM is a relatively portable, general purpose SVM implementation written in C++ that includes tools for training, as well as tools and code for classification.

FFSVM is a library that can load such models trained by libSVM's svm-train and offers a number of benefits:

Features

FFSVM

loads almost all libSVM types (C-SVC, ν-SVC, ε-SVR, ν-SVR) and kernels (linear, poly, RBF and sigmoid)
produces practically same classification results as libSVM
optimized for SIMD and can be mixed seamlessly with Rayon
written in 100% Rust, but can be loaded from any language (via FFI)
allocation-free during classification for dense SVMs
2.5x - 14x faster than libSVM for dense SVMs
extremely low classification times for small models (e.g., 128 SV, 16 dense attributes, linear ~ 500ns)
successfully used in Unity and VR projects (Windows & Android)
free of unsafe code ;)

FFSVM is not, however, a full libSVM replacement. Instead, it assumes you use svm-train at home (see Usage below), and ship a working model with your library or application.

Usage

If you have a libSVM model

In this example we assume you already have a libSVM that was trained with svm-train. If you haven't created a model yet, check out the FAQ on how to get started.

use ffsvm::*;
use std::convert::TryFrom;

fn main() -> Result<(), Error> {
    // Replace `SAMPLE_MODEL` with a `&str` to your model.
    let svm = DenseSVM::try_from(SAMPLE_MODEL)?;

    let mut problem = Problem::from(&svm);
    let features = problem.features();

    features[0] = 0.55838;
    features[1] = -0.157895;
    features[2] = 0.581292;
    features[3] = -0.221184;

    svm.predict_value(&mut problem)?;

    assert_eq!(problem.solution(), Solution::Label(42));

    Ok(())
}

Performance Tips

For a ~50% performance boost consider compiling your application with more aggressive CPU flags (e.g., export RUSTFLAGS="-C target-feature=+avx2" in case you run on a modern x86 CPU).
For a further x-fold performance increase, create a number of Problem structures, and process them with Rayon's par_iter.

Structs

DenseSVM	A SVM using SIMD intrinsics optimized for speed.
ModelFile	Parsing result of a model file used to instantiate a `DenseSVM` or `SparseSVM`.
Problem	A single problem a SVM should classify.
SparseSVM	A SVM optimized for large models with many empty attributes.

Enums

Error	Possible error types when classifying with one of the SVMs.
Solution	The result of a classification

Traits

Predict

Implemented by DenseSVM and SparseSVM to predict a Problem.

Type Definitions

DenseProblem	Problems produced for `DenseSVM`s.
SparseProblem	Problems produced for `SparseSVM`s.