How to Deploy Machine Learning Models (ft. Runway)

⚙️ Runway - MLOps made easy: bit.ly/mrxrunway
🛣 Full Stack Data science roadmap: shorturl.at/abiJY
📚 Designing Machine Learning Systems (by Chip Huyen) 👉 amzn.to/3Cajv0Y
👔 Need help with preparing for your next data science interview? Use referral code ‘thuvu’ to get 10% discount on any of the offerings on datainterview.com
🔑 TIMESTAMPS
================================
0:00 - Intro
0:51 - Model deployment explained
2:44 - What we are going to build
3:40 - Creating project, setting up dev. environment
4:33 - Link extension, data prep & model training
8:21 - Saving model to Runway
9:04 - Creating model API
10:21 - Making prediction requests using model API
10:56 - Redeploying model
12:07 - Deleting model and pipeline
12:32 - Conclusions
👩🏻‍💻 COURSES & RESOURCES
================================
📖 Google Advanced Data Analytics Certificate 👉 imp.i384100.net/anK9zZ
📖 Google Data Analytics Certificate 👉 imp.i384100.net/15v9y6
📖 Learn SQL Basics for Data Science Specialization 👉 imp.i384100.net/AovPnJ
📖 Excel Skills for Business 👉 coursera.pxf.io/doPaoy
📖 Machine Learning Specialization 👉 imp.i384100.net/RyjykN
📖 Data Visualization with Tableau Specialization 👉imp.i384100.net/n15XWR
📖 Deep Learning Specialization 👉 imp.i384100.net/zavBA0
📖 Mathematics for Machine Learning and Data Science Specialization 👉 imp.i384100.net/LXK0gj
📖 Applied Data Science with Python 👉 imp.i384100.net/gbxOqv
🙋🏻‍♀️ LET'S CONNECT!
================================
🤓 Join my Discord server: / discord
📩 Newsletter: thu-vu.ck.page/profile
✍ Medium: / vuthihienthu.ueb
🔗 All links: linktr.ee/thuvuanalytics
As a member of the Amazon and Coursera Affiliate Programs, I earn a commission from qualifying purchases on the links above. By using the links you help support this channel at no cost for you.
#machinelearning #datascience #ThuVu #dataanalytics

[@Thuvu5]

Check out Runway MLOps platform here: bit.ly/mrxrunway

[@rashedkhan4584]

Rgy

[@mostafaemad2235]

ضش😊

[@johankakou8045]

Thanks for the book ! I'm a new follower from ivory Coast and I really love the content !

[@MandarKarekar]

Great video , I will try runway , please keep making new videos like this introducing new tools

[@datanash8200]

Reminds me of the Steve Jobs quote "Real artists ship" so i guess our goal should be to "ship" i.e. actually deploy our model 😅 Ive also been guilty of leaving a few dusty jupyter notebooks 🤣

[@noorhassanwazir8133]

Your way of explanation is very well.

[@Thuvu5]

Thank you! I’m so glad to hear ☺️

[@ArsenBeglarian]

Big thankyous!! I got tired asking this question to ChatGPT 😂

[@gmostafaali]

Hi can you do more video about Runway

[@magnanimist]

I'm confused about their free tier pricing. Like... how far does the 125 credits in free tier actually get you for a small prediction model that isn't generating image or video content?

[@jxmusic357]

That's nice, but how to monitor accuracy and data drift

[@nguyencong9802]

Hi Thu Vu, thank you for sharing the platform. I'm curious that do you have in mind any platform or way which can be more friendly for normal users ( nontech users) to interact with model instead of using terminal? Many thanks!

[@renansk8flip]

The easiest way I can think to solve it is integrating the API into a simple HTML page who will receive the attributes needed for the model and will return visually to the user the result. Once the API is build, how friendly it will be for the user is the frontend developer. Its also possible if the API is build to accept of course, the attributes directly from the link, so for example, if you need 'age' and 'gender' attributes for a model, you can set up you API to receive that directly from the URL, so instead of sending a JSON file you send the info using the URL like ..//myapi.eg/?age=20&?gender=1

[@AIWizHash]

Amazing effort to explain deployment ❤

[@gviacava]

Can't get over GPT (-1). Great work!!! thank you!

[@Thuvu5]

Haha! I know right?! Thank you for checking out the video and commenting ❤️

[@MichealAngeloArts]

I found this video very informative! However, I noticed that there was no mention of how the Runway platform facilitates labeling data for use cases such as text classification. As someone who develops plenty of NLP models for my projects, in many cases I need to label the corpus myself. I use tools like Argilla to do so. However if I want to use Runway for MLOps, I would greatly appreciate it if you could provide more information on this topic. Could you please explain how one can label their data within the Runway platform, and if there are any built-in tools or features available that can help with this process?

[@Thuvu5]

Hi Ahmad, thanks for this question! I think currently this feature is not included in the Runway platform. The data labeling step is assumed to happen during the data preparation step in the pipeline already. Hope it's helpful 😊

[@atangbingana283]

Can you give me TLDR of what this model does ? why is this beneficial?

[@out-of-sight]

Hi, thanks for this beautiful content. Would you make a video like this with open source tools? Also I'm wondering about the mobile deployment of ai models. Thanks again. ❤

[@socode4714]

Cảm ơn chị ^^

[@doneaabohlassan1995]

It is payment? Or free

[@awedaoluwanifemi8024]

What is the difference with deploying this way and deploying with a framework like fastapi or flask?

[@tanthole0306]

Platforms like Runways are usually for people with little to no engineering skill to quickly deploy models. You can think of it as they create the FlaskApi code + the model, for you. Of course, if you want to dive into the engineering side and do that yourself, you will need to learn a framework (FastApi) and then set up everything. So basically these platforms save you time and effort to host a simple model, but if you want to do something big, usually you would build your own system. Hope this help❤

[@aryanpatel2114]

You’re the best Thu!

[@ganeshtiwari2550]

Sir need an advice I m from mumbai 29 age education gap like my graduation should complete in 2015 but completed in 2019 I m bachelor of banking n insurance student but I never work in that field sir.. I m from banking and insurance background I tryed many thing to do web development fire safety digital marketing n article writing facility management us taxatiin Can you plz suggest should I go in data analyst or sir Or not according to you.... You advice can change my life sir..

[@LukeBarousse]

print("hello data nerds")... the only TRUE way to test your code 🤓

[@Thuvu5]

Haha yesss, for sure!! I think I must have learned about this test from one of your videos 🤣🧐

[@debojitmandal8670]

@@Thuvu5 so true way is used to deploy ur jupyter notebook insted writing the code in vs code to deploy it

[@arokyajenson7878]

​@@Thuvu5😎😎

[@simonebrazzi287]

As always, great video😎 Is there a way to contact you? Like a discord channel

[@Thuvu5]

Thank you for watching, Simone! Yes you can join my Discord server, I often check out questions there 😊 discord.gg/Z53cQa3UFT

[@colonelchubbers]

Can someone elaborate on why / how models degrade over time?

[@RobWynn]

In a nutshell, because the real world is dynamic, not static. It is always changing. A ML model was trained on data at a particular point in time. As time goes on, things change - people change, user behavior changes, data collection methods change, technology changes, and so on. Features that were pretty important when the model was trained might not be as important next year, etc. Look at yourself as an example. Suppose a ML model was built when you were a child to predict the kind of TV shows/movies you would enjoy, using the data available from that time in your life. If that model were unchanged, would it still be accurate today? Probably not. You have changed. The world has changed. The available shows/movies have changed. That's why it's important to keep your models updated.

[@user-hx6xq9kn8o]

08:10

[@firefoxcarranza6432]

Que hermosa estás 😍

[@redwing6863]

momo

[@aj-lan284]

Are you on insta?