Presented By O’Reilly and Intel AI
Put AI to Work
April 29-30, 2018: Training
April 30-May 2, 2018: Tutorials & Conference
New York, NY

Getting up and running with TensorFlow

Amy Unruh (Google)
9:00am–12:30pm Monday, April 30, 2018
Implementing AI, Models and Methods
Location: Sutton South
Average rating: **...
(2.75, 4 ratings)

Who is this presentation for?

  • Developers interested in machine learning

Prerequisite knowledge

  • A basic understanding of Python
  • Familiarity with machine learning (useful but not required)

Materials or downloads needed in advance

What you'll learn

  • Learn how to build and deploy simple and complex models with TensorFlow


Amy Unruh walks you through the process of building a complete machine learning pipeline, covering ingest, exploration, training, evaluation, deployment, and prediction. Along the way, Amy explains how to explore and split large datasets correctly using BigQuery and Cloud Datalab.

Third-party libraries used:

  • TensorFlow—A machine learning library written in C++ with a robust Python API
  • pandas—An open source library providing high-performance, easy-to-use data structures, and data analysis tools for Python
  • The Jupyter Notebook—An open source web application that allows you to create and share documents that contain live code, equations, visualizations, and explanatory text


Machine learning and TensorFlow

  • What is ML, and why do we care?
  • Why is TensorFlow uniquely good or useful for ML?

A wide and deep thought experiment

Wide and deep code model

  • Input functions
  • Create, train, eval, predict loop
  • Run the code in Jupyter

Additional info

  • TensorBoard visualizations of the training and model graph
  • Limitations of this model

Diving into a lower level of TensorFlow

  • Using MNIST as a toy dataset to play with model structure
  • TensorFlow primitives

Creating a simple network by hand

  • Using the core TF libraries to create a model for solving MNIST
  • Tips and tricks for improving your model

Upgrading the model to a CNN (time permitting)

  • Creating CNN layers by hand
  • Available hyperparameters

Wrap-up and Q&A

  • Other models for other problem domains
  • Production environment considerations
  • Resources
Photo of Amy Unruh

Amy Unruh


Amy Unruh is a developer programs engineer for the Google Cloud Platform, where she focuses on machine learning and data analytics, as well as other Cloud Platform technologies. Amy has an academic background in CS/AI, and she’s worked at several startups as well as industrial R&D and published a book on App Engine.

Comments on this page are now closed.


Picture of Amy Unruh
04/29/2018 1:27pm EDT

hi Kate, that’s correct — CPU is fine.

Kate Baroni | ARCHITECT
04/29/2018 1:24pm EDT

Hi Amy: Just checking but we should be able to complete the labs using CPU (GPU not required), right? Thanks ~

Picture of Amy Unruh
04/29/2018 3:08am EDT

Ashley, If you don’t have python installed, conda might be easier ( ,
However, you’ll also be able to run almost everything using colab (, which is browser-based, so you can just go that route if you have installation issues.

ashley zhang | DATA SCIENTIST
04/28/2018 10:32pm EDT

I will bring a corporate laptop that has windows installed. How should I install virtualenv on top of it? Thanks.

Picture of Amy Unruh
04/26/2018 2:18pm EDT

Jim, that should be fine. You can just follow along, or maybe pair with someone nearby. All the code is on GitHub and you can try it later. You’ll also be able to run nearly everything on colab (

04/26/2018 1:28pm EDT

I will not have a laptop to bring. In your opinion, Will this tutorial be useful without hands on? Thanks.