Open Source Machine Learning from Google

We at “Whoop!” use advanced Machine Learning algorithms for predicting optimal Max-CPC-Bids for every Product based on the performance history of a retailers product range and many more different variables. So we are always curious about new things happening in the field of Machine Learning. Especially with the focus on improving a PPC tool.

 

Too long, didn’t read version:
• Google released TensorFlow as Open Source project under Apache 2 licence.
• TensorFlow makes Deep Learning available for everyone.
• Doesn’t really fit Big Data because the Open Source tool only supports single machine environments but multiple GPUs.

 

Machine Learning With Google

This week Google released its Deep Learning tool “TensorFlow” as Open Source project under the Apache 2 licence. The underlying algorithm was among other things used by Google to learn Google Photos what e.g. a cat looks like. Thus, typing a search query for “cat” can now learn to filter the pictures containing cats.

TensorFlow is written in C++ and a big part of the project was released under Open Source. It is also possible to create Python applications for TensorFlow. It can be used to build, train, test and deploy Neural Networks and other Machine Learning applications. Examples range from picture recognition via voice recognition to translation applications. Plus, Google promises easy use on mobile phones!

 

Should You Use it?

Here’s what our “Whoop!” Data Science team thinks about TensorFlow:

The first Google Deep Learning tool was DistBelief, developed back in 2011. As DistBelief was very complicated to configure and were coupled with Google’s infrastructure, DistBelief was designed to correct this disadvantage. The big cutback for this new Big Data tool is the limitation to single machine environments but one can still use multiple GPUs in the Open Source version1.

The new TensorFlow workflow is based on pretty computational flow charts, the first comparable interface which came to our mind is Azure ML2. You can see a quick example of the Google tool in reference3.

We are very excited about this new tool and we are looking forward to see some interesting Google Shopping or PPC tools based on TensorFlow in the near future!

Stay tuned for more interesting news from our “Whoop!” Data Science department!

 

Reference:
1 http://www.techtimes.com/articles/104924/20151110/google-open-sources-machine-learning-system-tensorflow-but-there-a-catch.htm
2 https://visualstudiomagazine.com/articles/2014/09/01/~/media/ECG/visualstudiomagazine/Images/2014/09/0914vsm_mccaffreyfig01.ashx
3 http://googleresearch.blogspot.de/2015/11/tensorflow-googles-latest-machine_9.html