To all my Machine Learning engineers, Ive been doing Frontend development for 6 years and I'm done. Wanting to get into machine learning because I've always loved data.

1. What is your day to day like?
2. Any advice for my learning journey?

Thank you🙏

Maths, lots of Maths

Our machine learning devs usually describe it as:

- determine algorithm
- testing algorithm, optimal data
- training
- analysis
- testing algorithm, realistic data
- back to 1

Rinse and repeat.

Having 6 years of FE in your baggage, with ML you could do wonders tweaking UIs to capture and analyse user behaviour.

@12bitfloat Music to my ears, statistics, probability, linear algebra, anything else?

@IntrusionCM Awesome, thanks for sharing!

@gosubinit Actually something I've never even thought of, thanks for pointing that out.

@tragicwhale as for advice...

From a manager perspective, thus a listener with no deep comprehension, it's dependent on how volatile your data is...

If the model can be easily trained and the data is consistent, happy ml devs.

If the model cannot be easy trained - well, a challenge.

If the model cannot be easy trained and the data is spooky dookey pukey shit so that each analysis phase becomes a crime scene with frustrating debugging of why it doesn't work aka where the deviations of optimal vs non optimal stem from.. Veeeeeryy mad and frustrated ml devs.

I bet it was web frontend.

Most of your time will go to data cleaning and preparation, less on algorithms or building and maintaining your model. Especially if you're dealing with unstructured data.

@Lensflare ha, ding ding. How could you have guessed so right.

get good with: statistics, probability, linear algebra, and calculus. Then see if you are interested.

If you have the money, try: https://nnfs.io/

And (this is free but get their book if you have it) https://www.fast.ai/

If you can follow along discovering the intricacies of the math involved in these two then you are set for more academic types. It is my impression that most of the tutorials out there focus on the academic advancements of the field of M.L rather than the practical one, as if you were supposed to come up with the next big CS. PHD paper to demonstrate something, but if you want real life value and work then go for those.

Still, you might not get an entrance to the field unless you: Have a PHD or advance scientific work, had built a framework of solved something big in Keggle, have a JARVIS type of system working for you altready

@bioDan we have data engineers to handle data prep, so that most things end up in a beautiful database.

However, and that is valid for everyone in the data stack:
- Users see *their* data as a good representation of *their* reality.
- Users do not care if their data is a good representation of the *actual* reality.

Thus, when you mix data silos you end up finding someone who is at the same time working, in vacation, have been fired last month and might have never been hired in the first place, depending on who you ask.
Multiverse of madness, yo.

So, @tragicwhale, remember that a dataset is only ever a good approximation of reality in a very specific context. Adapt your ML models, studies and conclusions accordingly.
And be prepared to argue reality vs data on a daily basis.

@AleCx04 this is such awesome information. I lived calculus when I was in college but never had to take statistics so I'll definitely hit that up. Thanks so much for all the information!

@tragicwhale glad I could help man! Have fun!