Install tinysvm with Homebrew, MacPorts

Project links

• Homepage
• Upstream docs
• Package manager page

Commands and aliases

• svm_classify
• svm_learn
• svm_model

Project history

The official TinySVM page describes it as an implementation of Support Vector Machines for pattern recognition, citing Vapnik's SVM work and positioning SVMs as then-new statistical learning algorithms for practical tasks such as text categorization and handwritten character recognition. Its own examples identify the package as 'TinySVM - tiny SVM package' and show a 2000 copyright line in the learner output.

The release notes show active development from at least January 2001 through August 2002. During that period TinySVM added support vector regression, Ruby bindings, RBF/Neural/ANOVA kernels, SWIG-based Perl and Ruby bindings, Python and Java interfaces, incremental training support, one-class SVM support, Mac OS X support, and Windows compiler support.

TinySVM was distributed in a very package-nerd friendly way for its era: source tarballs, Red Hat 6.x and 7.x RPM/SRPM directories, Windows binaries, and anonymous CVS checkout instructions from the author's site. The official page says development used CVS and invited users to join CVS-based development.

Adoption history

TinySVM's adoption appears to have been strongest among early SVM users who wanted a small Unix/Windows package with command-line tools and language bindings. The official feature list emphasizes sparse vectors, tens of thousands of training examples, hundreds of thousands of feature dimensions, LRU cache storage for Gram matrices, and optimizations inspired by SVM_light.

In modern package-manager culture it is mostly a preserved scientific-computing tool. The input metadata lists Homebrew and MacPorts packages, which suggests its current visibility is strongest among users maintaining old pipelines, comparing classic SVM implementations, or needing the exact svm_learn/svm_classify/svm_model command set.

How it is used

The command-line workflow is train, classify, and inspect: svm_learn reads training data and writes a model, svm_classify evaluates or interactively classifies test examples using that model, and svm_model displays model properties such as margin, VC dimension, and support-vector counts.

TinySVM accepts the same sparse training-data representation as SVM_light, using class labels followed by feature:value pairs. The official docs call out this format because it can represent large sparse feature vectors, an important fit for text and pattern-recognition workloads of the time.

Why package nerds care

TinySVM matters to package nerds as a compact fossil from the pre-scikit-learn era: a tarball/CVS-era ML library with CLI programs, RPMs, Windows binaries, and multiple scripting-language bindings. It is small enough to package, old enough to need compatibility care, and recognizable by its SVM_light-style data format.

Timeline

• 2000: Official command examples identify the package as TinySVM and show a 2000 copyright line.
• 2001-01-17: Version 0.02 added support vector regression and a Ruby module.
• 2001-09-03: RBF, Neural, and ANOVA kernels were added; SWIG-based bindings and Python/Java interfaces became available.
• 2001-12-07: Experimental one-class SVM support was added.
• 2002-03-08: Mac OS X support was added.
• 2002-08-20: TinySVM 0.09 was released with compiler and Windows build updates.

Related projects

• SVM_light is the closest implementation reference: TinySVM documents compatible sparse data representation and optimization algorithms stemming from SVM_light.
• SWIG is relevant because TinySVM used it to provide scripting-language bindings.
• LIBSVM is a related classic SVM package from the same general era, though not cited on the official TinySVM page.

Sources

• http://chasen.org/~taku/software/TinySVM
• source_facts.package-manager