RAPIDS

RAPIDS is an open source API and software suite, and machine learning framework with end to end collection of CUDA accelerated data science libraries. The software is designed to allow users to execute end-to-end data science and analytics pipelines entirely on GPUs, and utilize NVIDIA CUDA primitives for low-level compute optimization.

The software is meant to expose GPU parallelism and high-bandwidth memory speed through Python interfaces. It is meant to include support for multi-node, multi-GPU deployments, and to enable processing and training for large dataset sizes. RAPIDS was started from the Apache Arrow and GoAi projects based on a columnar, in-memory data structure that delivers data interchange.

RAPIDS provides native array_interface support so that data stored in Apache Arrow can be pushed to deep learning frameworks that accept or work with DLPack, Chainer, MXNet, and PyTorch. The software is meant to accelerate Python data science tool-chains with minimal code changes, and increase the speed of the iteration of machine learning models and deploying them at higher frequencies.

Projects and communities that RAPIDS is part of include blazingSQL, DASK, dmlc XGBOOST, Apache Spark, Plotly Dash, HPO, nuclio, Numba, sickit learn, and GoAi.

Some of the algorithm categories that RAPIDS supports include clustering, dimensionality reduction, linear models for regression or classification, nonlinear models for regression or classification, and time series algorithms. This includes density-based spatial clustering of applications with noise, K-means, uniform manifold approximation and projections, logistic regression, support vector machine classifiers, auto-regressive integrated moving averages, and more.