在内部它有一个指针传递机制，使它很容易传递numpy数组。 这种集成还使我们能够为任何可运行在JVM上的程序动态生成python绑定。 这意味着没有更多的过时的接口。在本演讲中，我们将展示其与keras的集成。
Deeplearning4j, the preferred JVM interface for JVM-based deep learning applications by KNIME, Weka, RapidMiner, and various Apache projects, has notoriously been a JVM-oriented play without a Python interface.
This is changing, and we are doing it right. Adam Gibson offers a high-level overview of jumpy, a better Python interface for deep learning applications, and explains why Spark’s Py4J interface for deep learning makes it impractical for deep learning applications. Jumpy uses Kinvy, the way to build Python apps with Android, to avoid network overhead. Jumpy also has a pointer passing mechanism, making it very easy to pass NumPy arrays around. This integration also allows users to dynamically generate Python bindings for anything runnable with a JVM, which means no more out-of-date interfaces.
Adam Gibson is the cofounder of Skymind, an enterprise deep learning and NLP firm, and creator of the distributed, open source frameworks Deeplearning4j and ND4J. Adam has taught machine learning at Zipfian Academy and is currently the deep learning specialist in residence at GalvanizeU. Adam has spoken at Hadoop Summit, OSCON, and Tech Planet in Seoul and is a coauthor of the forthcoming O’Reilly book Deep learning: A Practitioner’s Guide. Adam consults for hedge funds, Fortune 500 companies, and startups. He studied CS at Michigan Tech.
©2017, O'Reilly Media, Inc. • (800) 889-8969 or (707) 827-7019 • Monday-Friday 7:30am-5pm PT • All trademarks and registered trademarks appearing on oreilly.com are the property of their respective owners. • email@example.com