TensorFlow gains momentum: Twitter migrates to fan-favorite ML framework
Joining companies such as eBay and Google, Twitter now uses TensorFlow as its machine learning framework. TensorFlow continues to be a fan favorite in the framework wars and it’s no wonder why more and more companies are adopting the technology.
Twitter is settling its nest down in a new tree. TensorFlow is now Twitter’s framework of choice for machine learning. This open source software library has been making headlines left and right and it is continuously on GitHub’s trending repo list.
What prompted the switch? Twitter’s internal system for analyzing data is known as Deepbird and it has taken a long route to get to where it is now. Previously, Deepbird used Lua Torch but after the support for Lua Torch grew sparse the company migrated over to Python and explored their options.
It was then that they decided to go with TensorFlow. Why TensorFlow? The Twitter engineering blog states that “…TensorFlow had much better support for serving models in production.”
Better machine learning
Twitter already has reported that they are pleased with their choice to migrate and TensorFlow is here to stay. While working on DeepBird v2, teams have achieved higher productivity, easier access to machine learning, better inference performance, and improvement in model metrics. TensorFlow has provided twitter with a more simplified way to track metrics and enables their engineers to experiment more openly with data. With its ease of use, we hope to see some great things come out of this software adoption.
The Twitter tech blog lists a typical DeepBird v2 use case that TensorFlow will now be handling:
1) Frame the ML task: what are we optimizing, and what are the inputs and features?
2) Prepare datasets
3) Define a model suitable for the problem;
4) Write a training script to optimize a model and evaluate it on different datasets;
5) Define hyper-parameters and run the script on Aurora Mesos; and finally
6) Repeat from Steps 3 to 5 until the desired results are obtained.
How does the average Twitter user benefit from all these metrics? One of the uses for deep learning is how it improves Twitter timelines. Tweets are displayed to users based on relevancy, the Tweet’s author, and predicts how a user will react based on previous Twitter interactions. TensorFlow will now help with keeping your timeline shiny and full of quality Tweets that you’ll want to interact with. (Now you know who to blame after you’ve fallen down the black hole on your Twitter timeline and had hours past without you realizing it.)
Winning the popularity contest
TensorFlow continues to be a beloved machine learning software library. According to the 2018 StackOverflow Developer Survey, TensorFlow is the most loved framework with an impressive 73.5% of respondents praising it. The survey also states that it is one of the fastest growing technologies, and that couldn’t be more apparent.
There is now even a TensorFlow course available at Standford University that aims to teach the usage of TensorFlow in deep learning research, how to explore its functions, and how to build models for projects. Surely, this won’t the last course on this topic and as the machine learning buzz gathers momentum, there is no better time to become familiar with TensorFlow and how to utilize its architecture.
Who else is on the TensorFlow train? Among the familiar names are airbnb, eBay, nvidia, Google,Dropbox, Uber, DeepMind, and Lenovo. With such high-profile companies using the framework, it’s a safe prediction that this list will grow and others will start to adopt TensorFlow. Who will be next?