Why learn Keras? This neural network library is user-friendly and modular
Deep learning doesn’t have to stay science-fiction. When it comes to learning the latest ML, Keras is a great library for getting started and it can create some impressive feats. Where should you get started and why should you add it to your list of resolutions?
Today we are shining a spotlight on the Python library for deep learning: Keras. It prides itself on being user friendly and modular.
Keras runs on top of TensorFlow (or CNTK and Theano) and is a high-level neural networks API. Created by software engineer François Chollet, Keras allows for intuitive, fast experimentation.
Easier deep learning
Why should you use Keras?
We interviewed deep learning experts Uwe Friedrichsen and Shirin Glander about the past, present, and future of deep learning. When asked about their favorite tools, Uwe Friedrichsen responded:
I would perhaps add Keras, as it offers a uniform facade for various DL frameworks such as TensorFlow, CNTK, Theano and MXNet. You can use Keras to make DL locally on your computer but also as a frontend for the others of the big cloud providers.
- ML 101: Keras is one of the easier libraries to pick up for machine learning beginners. On GitHub, the documentation reads: “Keras is an API designed for human beings, not machines.” Simple APIs will help bring machine learning from sci-fi to your new reality. The “Getting Started” guide is even titled “30 seconds to Keras” and helps you keep it simple. By the end of the day, you’ll be stacking layers like a pro.
- Python-powered: Python is currently the most popular ML language. Keras is compatible with Python version 2.7 – version 3.6.
- TensorFlow support: TensorFlow just keeps gaining momentum. (Make your predictions now: Have you voted for TensorFlow as the dominant tech in 2019? I sure did.) In TensorFlow 1.4, one of the new biggest features was adding Keras to the core package. Read more about that in the documentation and see how to best utilize this partnership.
Need more convincing? Deep learning is in high demand and will continue to grow according to recent trends. Here’s a tweet from Keras creator François Chollet about its popularity in the tech startup sector:
What machine learning skillsets are in demand among tech startups? Here’s a quick crawl of the past 6 months of “Who is hiring?” job postings on Hacker News. pic.twitter.com/3MVFqaH0oI
— François Chollet (@fchollet) October 4, 2018
Begin your deep learning exploration
Thanks to Keras’ growing popularity, there is plenty to explore and learn. Here are some other tools and GitHub projects worth mentioning:
- Auto Keras: Built upon Keras, this neutral network library was created by DATA Lab for automated machine learning.
- Eclipse Picasso: This DNN visualization tool was developed to work with checkpointed Keras. (Checkpoints provided.)
- Deepjazz: Deep learning driven jazz generation using Keras and Theano. Built during a hackathon, this shows just how far your deep learning creations can go with the right creativity.
- Keras-vis: High level neutral network visualization toolkit
- Identifying dog breeds using Keras: The title speaks for itself! (Does this win the award for the cutest implementation of deep learning architectures?)
Is this library right for you? If not, what are the other alternatives?
- NeoPulse Modeling Language 2.0 has something to brag about. According to their creator DimensionalMechanics on the AWS marketplace: “NML can reduce code length by 85% compared to writing deep learning architectures in Python with Keras”.
- Gotta go fast? When it comes to speed, Keras lags behind the competition. However, speed is not everything. Its usability more than makes up for this.
- Torn between Keras and PyTorch for your first deep learning framework? This article by Piotr Migdal and Rafał Jakubanis explains the pros and in cons in depth. Each are good for different use cases and different kinds of users. The article also recommends choosing between these two frameworks, as other options are either no longer in active development, or lack the necessary flexibility.
Find Keras on GitHub and begin your machine learning quest. Who knows, it might one day become the industry standard.