and today I'm going to be talking about TensorFlow Extended,

also known as TFX, and how it helps you put

your amazing machine learning models into production.

This is the first episode in our five-part series

on real world machine learning which will help you

get up to speed on using TFX to create your own

production machine learning pipelines.

In today's episode, we'll be asking the question,

what exactly is this TFX thing anyway?

Let's find out.

♪ (upbeat music) ♪

When we think about ML, we usually only think

about the great models that we can now create.

After all, that's what all the research papers

are focused on.

But when we want to take that amazing model

and make it available to the world, we need to think about

all the things that a production solution requires.

So that's why we have TFX, to build production pipelines

so that we can offer our amazing models to the world.

Google created TFX because we needed it.

And there was nothing already available that could meet our needs.

Google does a ton of ML.

And not just Google but all of the alphabet companies.

There's ML in almost everything we do.

In fact, TFX wasn't the first ML pipeline framework

that Google created.

it evolved out of earlier attempts and is now the default framework

for the majority of Google's ML production solutions.

And now, Google has open sourcing TFX

and making it available to everyone.

And it's not just Google.

TFX has had deep impact on our partners,

including Twitter, Airbnb and PayPal.

As ML developers putting a model into production,

what do we need to think about?

First, when we start planning for developing an ML application,

we have all the normal ML things to think about.

That includes getting labeled data if we're doing supervised learning,

and making sure that our data set covers well the space

of possible inputs.

We also want to minimize the dimensionality of our feature set

while maximizing the predictive information it contains.

And we need to think about fairness.

And make sure that our application won't be unfairly biased.

We also need to consider rare conditions,

especially in applications like healthcare

where we might be making predictions

for conditions that only occur in rare, but important, situations.

And finally, we need to consider that this will be a living solution

that will evolve over time as new data flows in

and as conditions change and plan for life cycle management

of our data.

But in addition to all that, we need to remember that

we're putting a software application into production.

That means that we still have all the requirements that

any production software application has,

including scalability, consistency, modularity

and testability, as well as safety and security.

We're way beyond just training a model now.

By themselves, these are challenges

for any production software deployment.

And we can't forget about them just because we're doing ML.

How are we going to meet all these needs

and get our amazing new model into production?

We don't pretend to have all the answers.

This is an evolving field within the ML community

and we welcome contributions.

If you're interested in a more in-depth discussion

of the challenges of machine learning in production environments,

this is a great paper.

That's what TFX is all about.

TFX allows you to create production ML pipelines that include

many of the requirements for production software deployments

and best practices.

It starts with ingesting your data and flows through data validation,

feature engineering, training, evaluating and serving.

In addition to TensorFlow, itself, we've created libraries

for each of the major phases of an ML pipeline,

TensorFlow Data Validation, TensorFlow Transform

and TensorFlow Model Analysis.

TFX implements a series of pipeline components

which leverage these libraries,

which in this diagram are in orange, and allows you to create

your own components too.

To tie all this together, we created some horizontal layers

for things like pipeline storage, configuration and orchestration.

These layers are really important

for managing and optimizing your pipelines

and the applications that you run on them.

We'll be discussing those more in later episodes.

For now, that should give you an idea of what we're talking about

when we think about implementing a production ML pipeline with TFX.

In our next episode, we'll discuss how TFX pipelines actually work.

For more information on TFX, visit us at tensorflow.org/tfx

and don't forget to comment and like us below

and thanks for watching.

♪ (music) ♪