TensorFlow, Google's machine learning software library, has reached 1.0 status and now packs support for a wider selection of APIs for ingesting and analyzing Big Data. The tool's pre-1.0 life found it helping researchers, engineers, students, and others in examining information for patterns and insight. The full release makes the machine learning code even more powerful.
Google announced the changes to TensorFlow at its TensorFlow Developer Summit earlier this month. Google says TensorFlow 1.0 is faster, more flexible, and more production-ready than before.
On the speed front, a forthcoming update will let developers take advantage of a 7x speed increase in certain GPU configurations. Documentation from TensorFlow.org includes a handful of tips and tricks developers can put to use in their models to fine-tune performance. TensorFlow 1.0 also adds an experimental release of XLA to help developers target specific CPU and GPU arrays.
As for flexibility, Google says a new high-level API introduces a range of helpful modules (tf.layers, tf.metrics, and tf.losses). Further, the API tacks on a tf.keras module for compatibility with Keras.
Python developers will be happy to learn that TensorFlow 1.0 should deliver Python API stability, according to Google. Google changed the Python APIs so they more closely resemble NumPy. As a result, it should be easier for developers to implement new features without throwing a wrench in their code.
Beyond the big bullet points, TensorFlow 1.0 does include a handful of smaller feature improvements. For example, it includes a new debugger, which is a command-line interface and API for stripping bugs from your code. The update has new Android demonstrations for object recognition and camera-based stylization. There are some experimental APIs for Java and Go, as well as new compliance with PyPI pip packages.
"In just its first year, TensorFlow has helped [many people] make progress with everything from language translation to early detection of skin cancer and preventing blindness in diabetics," said Amy McDonald Sandjideh, Technical Program Manager, TensorFlow in a blog post. "We're incredibly grateful to the community of contributors, educators, and researchers who have made advances in deep learning available to everyone."