arctern-io / arctern

Apache License 2.0
103 stars 49 forks source link
geolocation geospatial gis gis-platform giscience gpu-acceleration gpu-programming

Arctern Docs

Arctern 中文文档

Overview

Arctern is a fast scalable spatial-temporal analytics framework.

Scalability is key to building productive data science pipelines. To address the scalability challenge, we launched Arctern, an open-source spatial-temporal analytic framework for boosting end-to-end data science performance. Arctern aims to improve scalability from two aspects:

Arctern's approach and current progress

We adopt GeoPandas‘s interface and plan to build the GeoDataFrame/GeoSeries that scale both up and out. On top of GeoDataFrame/GeoSeries, we will develop a consistent spatial-temporal algorithm set across execution environments.

We have now developed an efficient multi-thread GeoSeries implementation, and the distributed version is in progress. In the latest version 0.2.0, Arctern achieves 24x speedup against GeoPandas. Even under single-thread execution, Arctern outperforms GeoPandas 7x on average. The detailed evaluation results are illustrated in the figure below.

We are also conducting experimental GPU acceleration for spatial-temporal data analysis and rendering. By now Arctern provides six GPU-accelerated rendering methods and eight spatial-relation operations, which outperform their CPU-based counterparts with up to 36x speedup.

In the next few releases, our team will focus on: