Ruby's answer to scikit-learn

What it does Rumale is a Ruby machine learning library that implements common algorithms—linear models, trees, ensembles, multi-layer perceptrons, clustering, and dimensionality reduction—behind an interface deliberately modeled on Python’s scikit-learn. It loads LIBSVM-format datasets, supports cross-validation, and serializes trained models with Ruby’s Marshal.

The interesting bit The project does not try to out-Python Python; instead it offers Rubyists a native escape hatch from NumPy envy. Since version 2.0.0 it has switched to a custom Numo::NArray Alternative backend, and it can tap OpenBLAS or multi-core parallelism via optional companion gems when the workload demands it.

Key highlights

• Covers the usual suspects: SVM, logistic regression, random forests, gradient boosting, k-means, DBSCAN, PCA, t-SNE, and others
• Scikit-learn-style API with fit, transform, predict, and score methods
• Optional speedups through Numo::Linalg Alternative (OpenBLAS) and the Parallel gem for supported estimators
• Ecosystem extensions for LIBSVM/LIBLINEAR (Rumale::SVM) and torch.rb neural networks (Rumale::Torch)
• BSD-3-Clause licensed

Verdict Worth a look if you are committed to a Ruby stack and need standard ML primitives without bridging to Python. If you are already happy with scikit-learn, this is essentially a dialect translation you may not need.