https://kur.deepgram.com/images/logo.png

Kur: Descriptive Deep Learning

BUILD LICENSE PYTHON GITTER

Introduction

Welcome to Kur! You’ve found the future of deep learning!

  • Install Kur easily with pip install kur.
  • Design, train, and evaluate models without ever needing to code.
  • Describe your model with easily understandable concepts.
  • Quickly explore better versions of your model with the power of Jinja2 to automate Kurfile specification.
  • Supports Theano, TensorFlow, and PyTorch, and supports multi-GPU out-of-the-box.
  • Kur is open source and available at GitHub
  • COMING SOON: Share your models on KurHub, making it incredibly easy to collaborate on models and learn from others.

Here’s a good list to follow:

Like us? Share!

What is Kur?

Kur is a system for quickly building and applying state-of-the-art deep learning models to new and exciting problems. Kur was designed to appeal to the entire machine learning community, from novices to veterans. It uses specification files that are simple to read and author, meaning that you can get started building sophisticated models without ever needing to code. Even so, Kur exposes a friendly and extensible API to support advanced deep learning architectures or workflows. Excited? Jump straight into the Examples: In Depth.

How is Kur Different?

Kur represents a new paradigm for thinking about, building, and using state of the art deep learning models. Rather than thinking about your architecture as a series of tensor operations (tanh(W * x + b)) and getting lost in all the details, you can focus on describing the architecture you want to instantiate. Kur does the rest.

The Kur philosophy is that you should describe your model once and simply. Simple doesn’t mean brainless, nor does it imply that you are limited in what you can do. By “simple” we mean that models should be simple to use, simply to modify, and simple to share. A flexible, more general model is elegant. And this makes it easier to reuse in new contexts or to share with the community. Kur’s power lies in quickly making models that are both flexible and reusable.

Aside: Brief History Lesson

Decades ago, researchers wrote low-level code using highly optimized linear algebra libraries and ran the code on CPUs. After the rise of General Purpose Computing on GPUs (GPGPU), researchers modified their code to use CUDA or OpenCL. Although this code was functionally identical, GPU computing represented an incredible breakthrough in efficiency, as these new machine learning models could train and predict in fractions of the time compared to CPUs. Problematically, these programs were relatively hard-coded; exploring different hyperparameters or architectures typically required detailed knowledge of the code, and was fraught with ugly and bug-prone hacks.

Eventually, computer scientists began abstracting away the low-level, dirty details of highly-optimized CUDA code, and projects like Theano and TensorFlow were born. These tools are incredible in that they hide many of the implementation details of working with hardware (i.e., CPUs and GPUs), and instead expose higher-level tensor operations to the developer. Even then, the developer is forced to choose between building up higher-level abstractions of deep learning primitives, or devolving to the rigid or hacked models of earlier years. Projects like Keras and Lasagna emerged organically, driven by a need to more quickly and intuitively develop deep learning models. Their primary genius is in providing an API that maps to the way people actually think about the components of a deep learning network (e.g., as a “LSTM layer” rather than as a series of tensor operations).

Kur is the natural progression of these tools and abstrations. It allows you, the researcher, to get straight to the heart of deep learning: develop that awesome model you’ve been dreaming about in a few short lines. And best of all, you craft your model with high-level abstractions rather than having to think about annoying questions like:

  • Which language should I use?
  • Which backend is the best?
  • What if I want to quickly test different model configurations?