Although many data scientists would prefer to work entirely in Python, they may be compelled to use Spark or a lower-level programming language when scaling models and analyses or to effectively leverage task and data parallelism to iterate faster. To enable data scientists to use NumPy, a popular Python library, at scale, UC Berkeley researchers have OSS’ed NumS, which scales NumPy operations horizontally and provides task parallelism for these operations. NumS concurrently executes data structures called N-dimensional futures to effectively parallelize basic linear algebra operations on distributed memory systems.