Why this ML space is so confusing?
For SDE interviews, we still have an idea that DSA, LLD, HLD or for some companies they ask Java or Spring related questions..
But for ML why it's kinda uncertain ... I understand that it depends on roles... some requires statistics heavily, maths, deployment, core ML or DL algorithms..
But why interview process is not structured?
If you have any hints please drop here with specific company name..that will be of immense help!!
I find sde interviews confusing
For ml interviews just prepare the theory and do some major ml solution design. There are many ml/dl algorithms but with 1-2 months of prep it's easy, especially if you are using them in your work you'd get to know em naturally.
For design... I have heard people ask usually what they are working on.. irrespective of candidates' experience. Is it true?
Do not agree at all.. current ML space is extremely confusing..
some companies ask for research level experience with in-depth knowledge of all algorithms and their internal maths (which ideally should be the case) whereas some other companies focus on some particular practical exposure to certain areas.
For example if you have written llms in your cv, they will ask for random jargons and buzzwords that people write in LinkedIn to get attention and it's not that certain startups are doing so .Many big techs are also shifting towards that direction. This makes the ML interview process extremely unstructured.
And also not all companies require DSA rounds for ml, some require it and some focus on problem solving or a take home assignment. Some ask you to code k means from scratch and some may ask you to solve leetcode medium/hard problems. In rounds with hiring managers, one might ask problems related to recommendation systems whereas some may test your practical exposure to forecasting.
I don't know where you came to this conclusion from but it's exactly opposite
ML interviews are shit now a days!
Because the current situation is completely different, you need to clear DSA (Med & Hard as well) + ML questions(Can be anything in the space of ML from answering backprop to document chunking strategies)+ Design
And right now I am grinding DSA as I don’t want to f up any R1s.
And I have given 10+ interviews in the last 2months and everyone asked good level of DSA questions.
In one interview I was able to explain the transformer architecture but then question was on the time complexity of attention mechanism.
And on the other end someone ask questions on evals.
Now I am so confused what should you prep for these interviews as non of them follow the same pattern apart from the DSA questions.
On a side note, why is it so hard to get even a rejection email from companies, is there literally this much competition? For context I'm about to graduate and have a few projects up my sleeve, especially in predictive modelling and I can't find a job, placements this year were quite awful and just today as I was applying for jobs on LinkedIn, i saw a couple intern roles, so was editing the resume and as soon as I hit apply button I realised the roles were closed within 6 hours, this happened not once but twice on the same day
Totally agree based on my experience, most of the interviews focus on ML foundational concepts , but depending on the jd, they can go to forecasting or optimization, even if you nail all of them you never know they might throw a DSA problem on your way. Apart from that one definitely has to have mastery over python, sql and stats
What's the best way to prepare for an ML interview considering how vast the subject is any form of a roadmap or direction will be very helpful
ML interviews mostly test basics but yes they can become gnarly considering the vastness of the subject. For the most part you'll be fine with knowing how most popular algorithms work but yes they can expect code implementations of things like KNN and stuff and if they're being very picky go for stuff like ML system design
I believe that is how it should be. If it can be prepared for, then it's not a true test of aptitude. Just look at SDE interviews. DSA was supposed to test problem solving skills. But people just grind leetcode and memorize the most common patterns of problems to pass interviews.
