|TensorFlow For R|
|Written by Kay Ewbank|
|Wednesday, 14 February 2018|
The team at RStudio has created a set of R interfaces to TensorFlow, in recognition of the way the Google machine learning framework has become popular since it was open sourced two years ago.
RStudio is an integrated development environment for R that includes a code console, syntax-highlighting editor, and tools for plotting, history, debugging and workspace management. The new TensorFlow interface is made up of a suite of R packages, each offering a different interfact to TensorFlow for different tasks and levels of abstraction.
The new interfaces were announced at rstudio::conf, and the video of the session gives details of what the software is and what it does:
The packages start with an interface to Keras, a high-level neural networks API developed to enable users to carry out experiments quickly so they can get from an initial idea to a result with the least possible delay.
The second package is an interface to TensorFlow Estimators. This is a high-level API that offers implementations of a number of different model types including linear models and deep neural networks.
There's also an interface to the complete TensorFlow API, giving access to the set of Python modules that enable constructing and executing TensorFlow graphs.
The final interface is to the TensorFlow Dataset API. This can be used to create scalable input pipelines for TensorFlow models, covering tasks such as reading data from a variety of formats; transforming datasets in a number of ways including mapping arbitrary functions against them; and shuffling, batching, and repeating datasets over a number of epochs.
Alongside the interfaces, the TensorFlow for R package includes tools to help with training workflows, including real time feedback on training metrics within the RStudio IDE. The package also includes a number of ways to use GPUs in the cloud. This is necessary for many users because training neural networks is computationally demanding, and you'll get a much better performance if you're running the software on a machine that has a high-end NVIDIA GPU. If this isn't available, the package comes with an R interface to Google’s hosted machine learning engine, along with an Amazon EC2 image preconfigured with NVIDIA CUDA drivers, TensorFlow, the TensorFlow for R interface, as well as RStudio Server.
or email your comment to: firstname.lastname@example.org
|Last Updated ( Wednesday, 14 February 2018 )|