MLFlow Tutorial Part 1: Experiment Tracking

This tutorial will show you the basics of experiment tracking with MLFlow for TensorFlow, Sklearn, and other frameworks. Learn how to structurise your experiments, log everything you want, and save the best models for later use. You can code along or simply pull the notebook and read it at your own pace.
Links:
Github repo - github.com/aru...
FT Transformer Blog - / improving-tabtransform...

[@peegee101]

Love the haircut! Nicely done!

[@atruberts]

Thank you so much 😀

[@hndr91]

What a clear explanation, thanks man!

[@atruberts]

Glad it helped!

[@brahyamalmonteruiz9984]

excellent video, but the audio was too low

[@user-pp1kg5ts7q]

Great content man ..why am i not able to see the compare button in the ui ..using 2.7.1 version in windows chrome

[@AbdolaMike]

Loved the video. you got my sub!

[@atruberts]

Awesome, thank you! Sorry for taking so long to reply, I'm planning to be more active now 🤞

[@fkeb37e9w0]

I am running mlflow server with local host inside a vm and using the same as tracking uri, but when I do start_run() I get an error of 400 or 403. How do I resolve this.

[@NewDataTime]

Thank you

[@atruberts]

You're welcome, glad you enjoyed it!

[@miguelovallevillamil4953]

Amazing content

[@atruberts]

Thank you so much! More is coming soon :)

[@user-fb9zv9cf1s]

Good stuff!

[@trangmun2851]

Amazing thankyou so much!

[@DoubleJMc]

nice!

[@atruberts]

Thanks buddy! I hope it's useful 😉

[@Jerry-uc1pn]

Your audio is way way too soft. I could not hear anything.

[@AvijeetPandey]

Hi Antons, thanks for the helpful video. However, we only have access to R/Jupyter notebook inside a virtual environment at my workplace. Hence, all the ports are blocked. So, spawning that cool UI/backend server is not an option. Can we still use Mlflow tracking with just jupyter notebooks and local filesystem(maybe txt files)? Would be cool if you make a part 2 for the same.. Thanks

[@atruberts]

Hi Avijeet, I'm glad you liked it! Hmm, in theory everything that your experiments generate are stored locally, either in mlruns folder, or in the .db you've specified. You'll need to find out where exactly the parameters/metrics you need are stored, but once you have their location it'll be quite easy to access them and compare between each other.

[@atruberts]

Just out of curiosity, what does it tell you when you try to run mlflow ui command?

[@AvijeetPandey]

@@atruberts no issues when running the ui command, it gives me a URL with some port address, but unable to open that URL in browser due to IT policies, just like standard Web address block notice

[@AvijeetPandey]

@@atruberts OK, so SQLite backend is not necessary

[@user-lf8yv3ls7j]

I guess you can use ngrok to access your local port externally. For example, it works in Google colab to open mlflow ui. Although it seems less secure than using ui locally. You can find examples of using ngrok in zenml notebooks.