Data Science In Production Episode 2: Deploying Deep Learning models with TimTem
Earlier in December I was able to catch up with two incredible Data Scientists and Machine Learning Engineers Tim Scarfe (https://twitter.com/ecsquendor) and Tempest van Schaik (https://twitter.com/Dr_Tempest). Tim and Tempest both work for Microsoft as part of the Commercial Software Engineering team (CSE). They have worked together on a number of projects and have formed the super group TimTem (or TimTam as below).
I met with TimTem at the Microsoft offices in Paddington, London to record the 2nd podcast. We covered so much in the time we had. Episode 2 focuses on a project that TimTem delivered for Confused.com. The project was delivered over 3, 1 week sprints. In that time the team were able to design, build and deploy a deep neural network in to production. In this podcast we discuss the justification for the project, how the team were able to deliver so quickly, DataOps, Model Management and so much more!
In this episode, I ask the listeners “How are you testing your models”. If you have an interesting approach or story, please let me know in the comments below.
There were a lot of topics mentioned on this episode.
Future decoded. FD is a conference organised by Microsoft. In 2018 Tim and Tempest delivered their session “Building deployable and updateable deep-learning: the Confused.com journey“. The slides are available here.
Confused.com. Confused.com is an insurance comparison website, which is very popular in the UK
Francois Chollet - Author of THE book on Deep Learning - “Deep Learning with Python” (Manning publications).
Keras. Keras is an open source neural network library written in Python. It is capable of running on top of TensorFlow, Microsoft Cognitive Toolkit, or Theano
Fully connected neural net. Check out the following blog on different neural networks. Here
Model Management. You can read more about Model Management in out DataOps series here.
Docker & Kuberentes. Management and orchestration of models in containers.
Databricks. Databricks is a PAAS offering of Apache Spark in Azure. Feel free to check out my session at SQLBits in 2019 on Data Science in Databricks. Signup here