Future Proof DSJul 24, 2026

[MLOps Course 1] Setting Up Production-Ready ML Projects 🚀

56views

[MLOps Course 1] Setting Up Production-Ready ML Projects 🚀

Future Proof DS

Learn two efficient methods for setting up your machine learning projects for production. This tutorial covers initializing a project with uv and using a pre-structured repository template to streamline your development workflow.

[MLOps Course 1] Setting Up Production-Ready ML Projects 🚀

Future Proof DS

Learn two efficient methods for setting up your machine learning projects for production. This tutorial covers initializing a project with uv and using a pre-structured repository template to streamline your development workflow.

Chapters

00:00

Introduction

Overview of setting up machine learning projects for production environments.

00:19

Using UV

02:39

Cloning Template

03:34

Managing Git

04:18

Project Structure

06:04

Choosing Your Approach

Transcript

00:00

All right, now let me quickly show you how easy it is to start setting up your machine learning projects so they can reach production.

00:07

Now in the guide I show you two different ways of doing this.

00:09

The first one is using the package manager uv.

00:13

It gives you the initial structure and a lot of flexibility.

00:17

And the second one is using the GitHub repo that I provided, which you can clone and use as a sort of template.

00:23

Right, now let's start with the first method.

00:26

Now I'm here in terminal, all I need to do is to uv init, use the flag package and give this a name.

00:34

So in my case it's going to be my ML project.

00:39

Now within it we are going to initialize the project and with package we are telling uv that we want to use the standard structure, the source layout.

00:49

Okay?

00:49

And this is of course going to make it a lot easier to install and run my Python project.

00:54

Now let's run this.

00:57

Alright, so it was very easy.

00:58

Now we can go into the project which is cd my_ml_project, and then if I use the tree command, I can actually see the structure of this project.

01:09

So we can see that it created a pyproject file.

01:12

This is where I'm going to be managing my dependencies.

01:16

It also created a readme file as a placeholder, but most importantly it created the source directory and inside of it a directory called my ML project.

01:26

This is basically my package right now.

01:28

again, like I said, this is a very flexible way of initializing your projects.

01:33

It doesn't give you too much, but it gives you enough to get started.

01:36

And then along the way you can just set up whatever else you need.

01:39

All right, All right, now let's go over the second method of doing this, which is basically just cloning this repo and using it as a template.

01:48

Now in the readme here I provided more information on how you can do this.

01:52

You can actually use the GitHub CLI and just basically create a project based on this repo as a template.

02:01

Or you can just simply use git clone and give, this a name.

02:06

Alright, so let's try that out.

02:08

Alright, so we're back in terminal.

02:09

All I need to do is get clone and of course I need to give this a name.

02:17

So in my case I'm going to call it my ML project and then let's clone it.

02:23

All right, good.

02:24

Now let's go into that directory, my ML project, and then let's actually open this up in an IDE so it's a little bit easier for you to kind of see this.

02:36

So let's do Devin Desktop.

02:40

So now let me quickly walk you through the structure that you're getting here because this looks a lot more fuller than just using uv init package.

02:48

And the idea here is that if you know you're going to be building a machine learning model and you're going to eventually deploy it, then this is giving you everything that you need along the way.

02:59

And of course you can just modify it if something is a little bit off for your specific use case, but it already contains everything that you need.

03:07

all right, so it comes with a config directory.

03:10

This is going to be really useful as, you know, the project grows and you want to start tracking hyper parameters and you want to track, you know, the state of your model and so on.

03:20

So you can put all that information here.

03:23

It also comes with a data directory.

03:25

Now I'm giving you four levels for this.

03:27

We have raw, entering, process and prediction.

03:30

This is going to make sure that all of your data is tracked properly in the proper state.

03:37

But of course you can modify this if it doesn't work for you.

03:41

We also have a models directory.

03:43

This is where you're going to keep all of your pickle files.

03:46

You have your notebooks directory.

03:48

This is probably where you're going to start with the exploration, doing some eda.

03:54

You have the scripts directory.

03:56

This is going to eventually become your endpoint if you want to serve the model using, using FastAPI.

04:01

So you have evaluate train, you have the source directory.

04:06

This is your, basically your package, right?

04:09

And in here I'm giving you some placeholder scripts.

04:12

So we have one for data, so this is where the data processing happens.

04:17

We have one for feature engineering, one for prediction, and one for training.

04:21

Okay.

04:22

You also have a test directory.

04:23

This is where you're going to put all your tests.

04:26

And then of course we have a Python version here, file.

04:30

So this is for us, to pin on the specific Python version that we want to use.

04:36

And since we're using uv as our package manager, you can just do uv pin and then change the Python version to whatever fits your project.

04:45

I'm also giving you here some placeholders for, you know, as a docker file and Docker compose.

04:50

This is going to be really useful once you want to package your project and deploy it and you know, managing your packages.

05:01

It already comes with a couple of packages that are really common when you're building machine learning models.

05:07

And then a readme file.

05:10

Now, of course, this contains the information that comes from the template itself.

05:14

You can just delete that and then fill it in with whatever you want.

05:18

All right, so again, you have two options here.

05:21

You have a very flexible option with uv, or if you know you're going to be building a machine learning model and deploying it, this is a great option.

05:32

I would highly recommend that you start here because it also teaches you how to start setting up your structure properly right Now, one last thing that I want to mention, it might be useful to you.

05:42

Let's go back to the terminal.

05:44

one thing that happens when you clone that repo as a template is that you are also carrying all of the history of that repo.

05:52

And of course you don't want that, right?

05:54

You're starting fresh.

05:56

So one thing that you can do is if you're inside of that project, you can do rm -rf .git.

06:04

so it's going to delete the git history and then you can just do git init to start tracking.

06:12

So you're basically tracking fresh, right?

06:14

There is no history about how the template was created.

06:18

You're basically just starting your project fresh.

06:21

So just a quick tip in case you want to keep things clean and organized.

06:25

All right?

06:26

Hopefully it's helped.

[MLOps Course 1] Setting Up Production-Ready ML Projects 🚀 — Tella