Do all the things like ++ or -- rants, post your own rants, comment on others' rants and build your customized dev avatarSign Up
From the creators of devRant, Pipeless lets you power real-time personalized recommendations and activity feeds using a simple APILearn More
magicMirror10266301dData scientists are not coders. Sound weird, but that the way it is.
The problem is thier original field. Most of these guys and girls know math, and lots and lots of theory. they could not code a rest backend, or a simple Web Ui. They also don't care about operational costs - they want to use thier latest and greatest ML algorithems. In jupyter notebook. Using Matlab. Maybe Tensorflow.
I once interviewed for an ML-OPS position. That job was "take whatever crap is shitted out by the ML team, and put it in production. Somehow, nobody cares"
Start small. there are datasets you can download and play with. Maybe do an intro from Coursera, Or Udemy. Find a small personal project, and use it there. My most interesting personal ML project was automating a dumbass Flash game - find the thing on screen, mouse click the thing type of game. Fun times.
TwentyMinuteMan522301d@magicMirror my other problem is when I’m using tensorflow or any library where you’re just messing around with the hyper parameters and just pick an algorithm from the already provided toolbelt, I feel like I’m cheating or like I’m not actually learning things.
Like I wanted to write my own graph system for the neurons and manually write how the forward pass and backward pass would work instead of just using a lib which is pointless but at least it teaches me things.
But then since all these years I’ve been writing “purposeful” code I find it harder to write a shitty mirror of the same stuff someone implemented and optimized in tensorflow merely because I know it’s gonna be shitty and then my brain locks and doesn’t know what to do or procrastinates for months.
Behold The Reinvented Wheel! Now in a new improved Shape: Triangle!
Ok - ML is not the classifier algo. You should understand the process first: feature extraction, feature selection, training, n-fold testing and verification over a dataset, and then how to interpret the results. after you understand those subjects, and how they affect the results - then you can start writing your own training/classifier algo.
Hazarth7775301dRegarding your first problem. Just start using python, you will need to stop being a little elitist bitch about languages and delve into it like It's the first time you're learning a programming language, it will click with you soon.
Secondly, I had the same issue for a while, I recommend watching the excelent Neural Networks from Scratch by sentdex on YouTube. It's not long nor complete, but once you coded and understood NNs in raw Python it wont feel like cheating to use a Framework around all that boiler code
AvatarOfKaine3788300dDoes anyone else think machine learning is a bit of a fad that isn’t going any further ?
So I’m 20 years old. Got a decent job as software engineer with a really good pay and really want to break into machine learning.
Aside from the syntax when I’m watching tutorials or reading books, I see data scientists and mathematicians make design mistakes in their code and it hurts my eyes and triggers my ocd.
I need tips on how to put my mindset in a moldable state so I can judge less and learn more and absorb data. Like you know that philosophy that when u get old your brain can’t learn things as fast anymore? I feel like that’s already happening to me rn at the age of 20.