@Reply #2 - good point! Honestly, most of those 'zero to hero' courses are kind of a disaster. I had issues with those fast-track tutorials too, they're just not as good as expected when you actually need to debug a model or fix an error. Unfortunately, you can't really skip the foundations if you want to stop guessing. It's pretty unreliable to just tweak numbers without knowing the underlying mechanics. Quick question tho before you buy anything - are you building a simple content-based filter for the cameras or something with deep learning? That changes which math to focus on first. If you've got $60, try these:

• O'Reilly Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow 3rd Edition - usually around $52. It explains the calculus behind backprop while actually showing the code.
• No Starch Press Practical Statistics for Data Scientists 2nd Edition - about $35. Basically, just focus on how to read the symbols. Once the Greek letters don't scare you, the rest gets way easier.

> it feels like every time i think i understand what a model is doing i see some notation with capital Sigmas or partial derivatives and my brain just shuts off. Man, i totally get that. I remember trying to build my first image classifier years ago and feeling like a total fraud cuz i couldnt explain a simple dot product to save my life. I wasted way too much money on heavy university textbooks that just gathered dust before realizing you don't actually need a full degree for this. If you only have 60 bucks left, honestly skip the fancy paid bootcamps for now. You should grab Cambridge University Press Mathematics for Machine Learning Hardcover or just find the legal free PDF version online if you wanna save that cash for a hardware upgrade. In my experience, Linear Algebra is 80% of the battle because everything is just matrices and vectors under the hood. Once you get how those transform, those scary Sigmas just look like simple loops. Just focus on matrix multiplication and the basic chain rule for backprop... thatll get you way further than any generic tutorial.

Like someone mentioned, focusing on specific math instead of trying to learn everything at once is the only way to stay sane. i remember when i was first trying to build a custom ranking algorithm for a side project and i almost quit because every tutorial assumed i already knew multivariable calculus. i ended up buying a few different books to see what stuck, and some were way too academic for a dev just trying to get code running. here are a couple of reliable options that i found actually helpful without being boring:

• Manning Publications Math for Deep Learning by Ronald T. Kneusel - this one is great because it specifically targets the math you need for neural networks. it shows the notation and then immediately shows the python code, which helped me finally understand what those sigmas were doing in my loss functions.
• No Starch Press Bayesian Statistics the Fun Way by Will Kurt - since you are building a recommendation tool, you are gonna need to deal with probability. this book is super cheap (usually under thirty bucks) and it actually makes statistics click without feeling like a classroom lecture. honestly the manning book is probably your best bet for the budget. it feels more like a bridge for programmers than a math textbook. i spent way too long guessing at learning rates before i just sat down with that one for a few weekends. it definitely made my models more stable tho.

Re: "> it feels like every time i think..."

• I had issues with those generic guides too. Most cheap courses were not as good as expected. Honestly, O'Reilly Essential Math for Data Science by Thomas Nield was way better. It's around $40 for the 340-page guide. It focuses on the specific calculus needed for backprop, which finally made my recommender's loss function actually make sense for my camera site.