Introduction - Vertex AI for ML Operations

Summary

In this introductory video, Mike, a Google statistician, likens machine learning workflows to assembling jigsaw puzzles, emphasizing their uniqueness and importance in efficiency. He plans to share various workflows through a series of videos and a GitHub repository, featuring end-to-end machine learning projects using Google Cloud's Vertex AI. Mike encourages viewers to explore all videos for new insights and suggests learning resources for those interested in machine learning, inviting feedback and collaboration to improve the shared knowledge base.

Takeaways

• 🧩 The video uses the analogy of jigsaw puzzles to discuss the uniqueness and personalization in machine learning workflows.
• 🔍 Mike, a Google statistician, aims to share and explore different machine learning workflows in this series of videos.
• 📚 He emphasizes the importance of learning from others' workflows to improve one's own efficiency in machine learning projects.
• 🌐 The series will cover workflows using Google Cloud and Vertex AI, focusing on their integration and application in end-to-end machine learning projects.
• 📘 Content will be based on a GitHub repository containing Jupyter notebooks, which is a common tool in the machine learning community.
• 🛠️ The series will guide viewers on setting up a Google Cloud environment, sourcing data, and using various machine learning methods including AutoML, custom training, and BigQuery ML.
• 🎓 Mike recommends resources for learning machine learning, including Google's Machine Learning Crash Course, Andrew Ng's course on Coursera, and Grant Sanderson's (3Blue1Brown) YouTube playlists.
• 🔑 The video series is designed to be modular, allowing viewers to skip around based on their interests, but Mike encourages watching all to gain a comprehensive understanding.
• 🤖 The series will not delve into the low-level details of machine learning algorithms but will focus on connecting process steps to form efficient workflows.
• 💡 Mike invites feedback and suggestions for additional workflows or improvements, leveraging the GitHub repository's issue feature for community collaboration.
• 👍 He concludes by encouraging viewers to like, subscribe, and comment to show support and contribute to the community.

Q & A

What is the main theme of the video series presented by Mike?

The main theme of the video series is sharing machine learning workflows using Google Cloud and Vertex AI, with a focus on the unique approaches people take to solve problems, similar to how different people approach jigsaw puzzles.

Why does Mike compare machine learning workflows to jigsaw puzzles?

Mike compares machine learning workflows to jigsaw puzzles to illustrate the unique and personalized nature of problem-solving approaches in both scenarios, and to emphasize the value of learning from different techniques.

What does Mike find most interesting about observing others build puzzles?

What Mike finds most interesting is observing the different techniques used by others, which can inspire him to try new methods and potentially become more efficient at tackling puzzles.

What is the purpose of the GitHub repository mentioned in the script?

The GitHub repository serves as a collection of Jupyter notebooks that demonstrate end-to-end machine learning workflows using Google Cloud and Vertex AI, allowing viewers to follow along and learn from the examples provided.

Why does Mike emphasize the importance of learning from others' workflows?

Mike emphasizes the importance of learning from others' workflows because they reflect the accumulated knowledge and efficiency strategies of individuals, which can help one quickly improve their own practices in the field of machine learning.

What are the different methods Mike plans to cover in the video series?

Mike plans to cover methods such as no-clicking no-coding model building and deployment, custom training, using BigQuery's built-in machine learning, and working with TensorFlow, among others.

What does Mike suggest for someone who wants to learn more about machine learning?

Mike suggests starting with a high-level overview from the Google Machine Learning Crash Course, then watching Grant Sanderson's video series on neural networks, and finally taking the Machine Learning course from Stanford on Coursera for a deeper understanding of the fundamentals.

What is the structure of the video series according to the transcript?

The video series is structured into multiple parts, starting with setting up the environment, followed by data sourcing, and then covering various methods of model training and deployment, including no-clicking no-coding, custom training, and using different tools like TensorFlow and BigQuery ML.

How does Mike plan to make the video series accessible and modular?

Mike plans to make the video series modular by ensuring each video starts with prerequisites, allowing viewers to skip around or focus on specific parts that interest them, while still being able to follow along with the workflow.

What is the role of the GitHub repository in relation to the video series?

The GitHub repository is where the Jupyter notebooks corresponding to each video are hosted, providing a practical, interactive way for viewers to engage with the workflows demonstrated in the series.

How does Mike encourage viewer engagement and feedback?

Mike encourages viewer engagement by asking viewers to like, subscribe, and comment on the videos, as well as to use the GitHub repository's issues feature to provide feedback, suggest improvements, or start discussions.

Outlines

00:00

🧩 Introduction to Machine Learning Workflows

In this introductory paragraph, Mike, a Googler and statistician, sets the stage for a series of videos on machine learning workflows. He uses the analogy of assembling jigsaw puzzles to illustrate the unique and personal nature of workflows in machine learning. He explains that, much like puzzle assembly techniques vary among individuals, so do the workflows in machine learning. Mike's passion for learning and sharing is evident as he invites viewers into his office and introduces the concept of exploring different techniques to improve efficiency. The video series aims to share workflows observed in the field, with the goal of helping viewers incorporate new methods into their future projects. The content will be hosted on a GitHub repository, featuring Jupyter notebooks as a common platform for machine learning work.

05:00

🛠️ Setting Up the Machine Learning Environment

Mike outlines the structure of the video series, which begins with setting up a Google Cloud environment in the first video. He plans to introduce a data source and discuss the selection process in the following videos. The series will cover various methods of training and deploying machine learning models, including no-click procedures, coding with Python, and higher-level orchestration. Mike emphasizes the importance of understanding the core components of Vertex AI and Google Cloud Platform to tailor workflows to individual needs. He also addresses common questions, such as the scope of the series, the order of watching the videos, and how to learn machine learning. He suggests that while the series won't delve into the minutiae of every machine learning method, it will focus on connecting process steps to form efficient workflows.

10:02

📚 Recommended Learning Resources for Machine Learning

In this paragraph, Mike shares his favorite resources for learning machine learning, acknowledging that the series is not a comprehensive tutorial on the subject. He recommends starting with Google's 'Machine Learning Crash Course' for an overview of terminology and problem framing. Next, he suggests the 'Machine Learning' course by Andrew Ng on Coursera for a deeper understanding of algorithms. Lastly, he points to YouTube playlists by Grant Sanderson of '3Blue1Brown' for visually engaging explanations of neural networks and foundational math concepts. Mike encourages viewers to engage with the content, provide feedback, and use the GitHub repository for ongoing collaboration and updates to the shared workflows.

Keywords

💡Jigsaw Puzzles

Jigsaw puzzles are a metaphor used in the video to illustrate the process of building machine learning workflows. They represent the complex and varied approaches one might take to solve a problem, much like assembling a puzzle from many pieces. In the script, the speaker compares different techniques of solving jigsaw puzzles within his family to the diverse workflows in machine learning, emphasizing the uniqueness and personalization of each approach.

💡Machine Learning

Machine learning is a core concept in the video, referring to the field of study that gives computers the ability to learn and improve from experience without being explicitly programmed. The video discusses machine learning workflows, which are the processes and steps involved in creating machine learning models. The script mentions various machine learning techniques and tools, such as Automl, custom training, and BigQuery's machine learning capabilities.

💡Workflows

Workflows in the context of this video are the sequences of steps, tools, and methods used to accomplish a machine learning project. The speaker believes that workflows are as unique as fingerprints, indicating the individuality and personal touch each professional brings to their work. The video aims to share different workflows to inspire viewers to consider new techniques for their own projects.

💡Google Cloud

Google Cloud is a suite of cloud computing services offered by Google. It is mentioned in the script as the platform on which the machine learning workflows are demonstrated. The video series will guide viewers on setting up a Google Cloud environment and using its services like Vertex AI and BigQuery for machine learning tasks.

💡Vertex AI

Vertex AI is a Google Cloud service for building and deploying machine learning models. The video script discusses using Vertex AI for various machine learning workflows, including automated procedures, custom training, and deployment of models. It is positioned as a key component in the Google Cloud ecosystem for machine learning.

💡Automated Procedures

Automated procedures refer to processes that can be executed without manual intervention. In the script, the speaker mentions using automated procedures like AutoML to train and deploy models, which simplifies the machine learning process by reducing the need for detailed coding or configuration.

💡Custom Training

Custom training is a process in machine learning where models are trained using specific data sets and configurations tailored to the user's needs. The video script mentions custom training as one of the methods to be explored, allowing for more control over the machine learning process compared to automated procedures.

💡BigQuery

BigQuery is a fully-managed, serverless data warehouse that enables super-fast SQL queries using the processing power of Google's infrastructure. The script mentions BigQuery's built-in machine learning capabilities, which allow for data analysis and model training directly within the data warehouse.

💡TensorFlow

TensorFlow is an open-source software library for dataflow and differentiable programming across a range of tasks, and it is also a popular framework for machine learning. The video script includes TensorFlow as one of the tools to be used in custom modeling workflows, highlighting its flexibility and wide usage in the field.

💡GitHub Repository

A GitHub repository is a location where projects are stored and version-controlled using Git. In the script, the speaker mentions a GitHub repository containing Jupyter notebooks that will be used to demonstrate the machine learning workflows. This repository serves as a practical resource for viewers to follow along with the video series.

💡Jupyter Notebooks

Jupyter Notebooks are open-source web applications that allow creation and sharing of documents containing live code, equations, visualizations, and narrative text. The video script refers to Jupyter Notebooks as the format for the content in the GitHub repository, which will be used to demonstrate and share machine learning workflows.

💡Orchestration

Orchestration in the context of this video refers to the automation of complex processes into a coherent sequence. The speaker mentions 'bottling up' code into a higher level of abstraction called orchestration, which streamlines the workflow by automating repetitive tasks, making the process more efficient.

Highlights

Mike, a Googler and statistician, shares his passion for learning and sharing in the context of machine learning workflows.

Jigsaw puzzles are used as a metaphor for the diverse approaches to solving problems in machine learning.

Workflows in machine learning are as unique as fingerprints, reflecting individual experiences and efficiencies.

The importance of learning from others' workflows to improve one's own efficiency in machine learning projects.

Introduction of a series of videos focusing on sharing workflows in machine learning using Google Cloud and Vertex AI.

The video series will cover end-to-end machine learning projects, showing different techniques and tools.

The series will be structured to allow viewers to skip around based on their interests.

GitHub repository containing Jupyter notebooks will be the central resource for the video series.

Explanation of how to set up a Google Cloud environment for machine learning.

Discussion on selecting a data source and making it available for various training methods.

Introduction to using automated procedures like AutoML for training and deploying models.

Use of custom training, BigQuery's built-in machine learning, and TensorFlow in the series.

Visual aids will be provided to help explain the workflows and processes in the videos.

The series will not cover very low-level details to maintain focus on connecting process steps in workflows.

Viewers are encouraged to watch all videos for a comprehensive understanding, but each video will have prerequisites listed.

Mike's recommendations for learning machine learning, including courses and YouTube playlists.

Invitation for feedback and collaboration through GitHub issues to improve and expand the repository.

Emphasis on the collaborative nature of AI and ML, aiming to make the field more approachable and well-connected.

Transcripts

welcome i'm mike a googler a

statistician and i'm passionate about

learning and sharing today i'm a friend

and i welcome you

in

to my office

here we are

i want to start this series with a

little story

it involves these you probably recognize

these these are jigsaw puzzle pieces uh

jigsaw puzzles are a big part of my

family these two pieces go to

this gigantic puzzle that we're working

on right now has

more than 2 000 pieces

and i bring this up for a reason

i find that there's a lot of things that

we do in common when different people

build jigsaw puzzles maybe we aim for

the corner pieces

maybe we fill in the edges because the

pieces are a little easier to identify

they have a flat side depending on the

cut of the puzzle

and it's a little easier to frame and

see the size of what you're about to

work on but when you go to the middle my

preferred method is looking for objects

that have like clear lines and then i

can pick those pieces uh

my wife she likes to pick ones where

like sections with a color that's easy

to identify amongst all the pieces

i find that when i work with my mom

she takes the approach of kind of inside

out she looks for features and she

builds upon those and then connects them

to adjacent features

but what's most interesting about

watching family members and others build

puzzles is i see techniques that aren't

my own

and then when i get a new puzzle i might

consider trying a different technique

that might help me or make me more

efficient at that particular puzzle

why are we talking about puzzles you

clicked on a link that's related to

machine learning and google cloud with

vertex ai well here's why

i believe

workflows are the most important part of

how we do

our work and in machine learning

workflows are almost as unique as a

fingerprint

my favorite thing to do when i go to

conferences or meet new people who are

in the same field is ask them

how do you do your work and why do you

do it that way i remember being very

very young as a statistician and going

to

uh the joint statistical meeting the

largest gathering of statisticians in

the world and whenever i'd meet someone

the first thing i'd say is what software

do you use how do you use it why do you

use it that way and i would hopefully

get them to show me on their computer at

the time

how they did their job and that made me

an infinitely better statistician very

very quickly because

the workflows are like the mirrors of

what they've learned throughout their

career to make themselves efficient and

i can immediately assimilate and decide

do i want that do i want to take parts

of that do i want to consider that when

a new project comes up where that might

fit better than my typical way and that

is what this series of videos is going

to be about it's going to be about

sharing workflows

that

i have seen with customers with friends

my own curiosity

and i'm going to put them together in

end-to-end pieces that show like a

machine learning project

and share it with you i'm going to try

to make them independent so that you can

skip around or only visit the ones that

seem to match what you're interested in

but i encourage you to try visiting all

of these and see

maybe it offers something that you can

incorporate in one of your future

projects

so let's take a look at a few things

here

this is all going to be based on a

github repository i put it in the lower

screen here

um

let's actually make that bigger so we

can all see it

all right

in this

series it's a github repo made up of

jupiter notebooks a common way that we

work in machine learning

you can visit these notebooks directly

in this repository it's linked in the

description of the video if you ended up

at the video first

maybe you ended up at the repository and

you clicked the video here and that's

the one you're watching

within here i'm going to

tell you how to set up a google cloud

environment that's the next video i'm

going to

bring in a data source and make it

available for all the different training

methods that we're going to use

and i'll talk about why i picked that

one and how others might work that'll be

the next video after that one

and then we will start going through

using automated procedures like automl

to train a model and deploy a model

we'll use custom training

we'll even use bigquery's built-in

machine learning for part of this we'll

use tensorflow

as you go through these i've put up some

visuals that i'm going to use in the

videos to help explain what i'm doing uh

one of my favorites is this one i'm

going to zoom in a little bit here and

let's take a look

let's see

each of these columns is one of these

workflows uh video zero that's setting

up the environment one is the data

source and then this starts with two

there are two uh three videos in that

are labeled two a b and c

the first is going to use no clicking no

coding it's it are all clicking no

coding we're gonna go to a console and

we're gonna build a model and deploy a

model and get predictions all directly

through the web browser in the second

one we're going to replicate the same

thing but we're going to actually write

code from python to orchestrate all the

same things that we clicked with in the

first one and the third

we're going to bottle up that code into

a higher level abstraction called

orchestration

and we're going to kind of like build a

treadmill or a people mover at the

airport to just automatically move

through a lot of the process steps

then videos three those have two a and b

we're going to look at completely

working inside bigquery and then in the

second of those videos we're gonna

extract the model from bigquery and

bring it to vertex and deploy it

then

with four and five and on we're gonna

take custom modeling we're gonna take

tensorflow we're gonna take scikit-learn

we're gonna take

a number of other tools and we're going

to

train models and deploy models but we're

gonna use the core components of vertex

ai and google cloud platform to show

that it enables whatever fingerprint of

a workflow you want

before we get going further

there's probably some common questions

that people ask at this point so let's

do some q a

how about this question first hi mike

what is not covered in these workflows

what's not covered is going to be

uh very low level details i love details

um frequent feedback i get is maybe back

off of the detail and you're probably

going to give me that feedback as well

what i want to do is take us to the

point where all the process steps

connect and focus on the details of

connecting those to create workflows but

once you're inside of a process step

like machine learning we might not go

into all the different methods and all

the different architectures and all the

details of how to perfect an

architecture for a particular data type

we'll back off of that maybe save that

for a different video series so that we

can focus in our workflows here

do i have to watch all the videos does

order matter

uh i would encourage you to watch them

all because that's how we assimilate new

ideas you might be bored in a few maybe

it's like not interesting to your

particular field or how you work but

it's still exposure to something that

matters to someone somewhere that's why

i put it here

it just

maybe exposes you to a new tool that

you'll use but if you don't if you go to

any additional video in the series the

first thing i'll start with is

prerequisites which previous ones are

previous notebooks with videos

you might need to get to that point and

run it

but i've always made the path to each

one as short as possible

how do i learn machine learning

this is my favorite question uh

unfortunately it's not what this video

series is about but i have an answer for

you at least my answer

this is the best part about living right

now is if you ask this question 20 years

ago

both the field wasn't quite as mature

but the availability of great content

like youtube and coursera and

pluralsight and a million other sources

they just weren't out there and now

they're everywhere so i'm going to give

you my favorites let's jump over to the

screen

and let's scroll down in the repository

to a section i added in to the readme

called exactly this learning machine

learning

okay so it's made up of three sources

again opinionated

but these are my favorites and i've

shared them with people who've given me

feedback who are like mike this really

really helped now you're gonna you may

be coming from i've had linear algebra

in calculus and i get all this stuff i

know how it works

still peruse these because if you're

like me you you love things that confirm

your knowledge and these will do that

and they will also give you great ways

of showing how to explain it to others

because these are master teachers in

here way better than me

um the first is a course from google

called the machine learning crash course

this is great for terminology in

explaining how to identify an ml problem

how to frame that problem exposes at a

high level to a lot of different methods

for supervised and unsupervised learning

it references tensorflow apis

so it gives you ability to quickly go

hands-on with some maybe example data

it doesn't go into a lot of details

about the mechanics of what's happening

underneath

and that's why we go to number two on

this list and that's uh

my favorite course i think somebody told

me this is how coursera started was this

course i hope that's true but this is a

andrewing

it's offered by stanford on coursera and

it's just called machine learning and it

takes you through a few weeks of

taking a couple of algorithms and hand

coding like building them

you will understand how these work after

doing this yeah you might complain about

software choice uh octave you you might

not like that part but it's perfect just

do it exactly the way they explain it

and

use it just as a chance to just absorb

information about how things work

now

the third thing i put on here is a list

of

three youtube playlists from youtuber

grant sanderson with channel called

three blue one brown this

grant is a master explainer of very

advanced information these playlists are

my favorite i rewatched them

occasionally the first is a neural net

playlist it has four videos it'll take

you through what a neural net is a

simple one

it will create it he does everything

very visually has his own package to do

those

he even gets into the fourth video the

calculus of back propagation

that is the moment where the light bulb

comes on for most people like when they

understand

how the neural net actually learns the

gradient descent the minimizing of a

loss function how does that

mathematically work it's beautiful

now as he gets into that video uh

you might be begging for a refresher if

you're not up to speed uh relevant

currently relevant on linear algebra and

calculus he's also got two great

playlists for those and i put those in

here

now the order of these i recommend

getting the high level from the google

crash course machine learning crash

course

going and watching the video series with

grant

then going and getting the fundamentals

from the machine learning course

stanford via coursera

and then re-watching those videos from

grant the light bulb just comes on like

i smile when i watch these i wish i was

as brilliant as these people at just

teaching and explaining information

i think you'll like it uh and if you

have a better answer give that to me in

the comments or an email because

i love this stuff i love learning but i

also like watching how other people

teach so i can get better at it

all right let's jump back to q a i think

we have one more question

what if i want more detail or i have

questions

uh

this is great this brings us kind of to

the the wrap up here

uh you're gonna have

a feeling after you watch this video if

you like it hit the like button if you

dislike it

hit the dislike button uh try to hit the

like button

that's like giving me a cookie gives me

encouragement to keep making these

videos and adding more notebooks to this

repository um

[Music]

additional thing if you want more

content like this hit the subscribe

button that lets me know hey people are

waiting on this kind of stuff like make

some more

and i'll even reference others if i see

subscribers hitting this channel

if you want to be alerted when i launch

a new video hit the bell also

now the next

every youtuber says those things

i'm not a youtuber

did this help you

that's like giving a cookie not just to

me but to others who are watching the

video leave a comment say hey this video

helped me this way a little word of

affirmation it gives encouragement to

others who are watching

but the third thing maybe the most

important part is you have feedback hey

here's an additional workflow that i

like i would love for you to include

something like that

i have an improvement like something

you've done could be better i have a

correction something you did wasn't

quite right or you misspoke okay give me

that as well but here's a great thing

since we're using a github repository go

over there click the issues button at

the top start a new issue we'll have a

conversation maybe we'll collaborate and

create but that way the repository can

be updated way faster than i can remake

one of these videos and that will make

the information relevant to everyone

almost immediately

and with that i just want to say thank

you for hanging in here for this intro

video thank you for your time thank you

for your feedback that's going to be

coming let's work together to make the

practice of ai and ml more collaborative

more accurate

more approachable

to a wider and more well-connected

audience have a great day