Cuml Machine Learning

This wraps NVIDIA RAPIDS cuML for GPU-accelerated machine learning with a scikit-learn compatible API. You get the same RandomForest, KMeans, PCA, and regression models you already know, but they run on GPU and can handle datasets with 10K+ rows dramatically faster. The skill includes proper initialization boilerplate that smoke-tests GPU availability and falls back to scikit-learn if needed, plus helpers for the cuDF to pandas conversion dance. One thing to remember is cuML wants float32 dense arrays and doesn't support sparse matrices, so you'll need to cast and densify your data first. Good for classification, regression, clustering, and dimensionality reduction when you have real data volume.