News for hackers. Stuff that matters.

Where to learn the math for data science?

Hope you all are doing great. I would love to do lower-end kaggle competitions in the coming months. In high school we have not learnt any advanced math yet. I know python well. Where would you recommend to learn the math essentials for starting to learn machine learning/data science from like basics?
( 9 comments | 15 points | on HN )
Re:Where to learn the math for data science? by logicslave on Wednesday July 29, 2020 @02:02PM (#23989714)
Don't aimlessly learn math. Even if its from a statistics or machine learning book. Choose a well regarded model and then study all math that it takes to understand and use the model well. Learn nothing outside of this. Rinse and repeat.
Re:Where to learn the math for data science? by marapuru on Wednesday July 29, 2020 @04:05PM (#23990969)
I'm currently in a data science traineeship and got this exact advice from my teachers.

We had to go through the following article two, three, four times and recreate it ourselves and then explain all the principles to our fellow students. This helped me a lot in understanding the math behind it.

https://towardsdatascience.com/netflix-and-chill-building-a-...

PS. I'm a sucker when it comes to math. When I got the infamous math questions with metaphorical elephants on weight scales I couldn't stop but wonder why the elephant was on the scale.

Re:Where to learn the math for data science? by ketanmaheshwari on Wednesday July 29, 2020 @01:31PM (#23989400)
Since you know Python, this book called Think Stats would be a good start, I think. A free PDF is available here: https://greenteapress.com/wp/think-stats-2e/
Re:Where to learn the math for data science? by usgroup on Wednesday July 29, 2020 @01:51PM (#23989612)
Not sure you need any maths for Kaggle competitions. Learn XGBoost and random forests for regression and classifications and learn how to evaluate the models and some of the tricks for coercing different types of data into them for good peformance. Then curve fit on various data to your heart's content.

It'll take ages before you need maths more complicated than arithmetic.

Taking a stats course won't help you at all. The way canned ML algorithms work has nothing whatsoever to do with data generating models which is what stats is about.

You'll start to need maths when it comes to dimensionality reduction and embedding and that'll come in the form of 95% linear algebra and 5% calculus. Still no stats though.

Wayyyy down the line if you get into the information theoretic view, or pick up the fetish of justifying everything via a Bayesian methodology then you'll need lots of stats.

Re:Where to learn the math for data science? by needsbetternews on Thursday July 30, 2020 @03:15PM (#24001864)
This course from carnegie mellon university has recorded lectures you can watch and extensive notes which will give you a crash course in the applied math needed like a linear algebra crash course, probability http://www.datasciencecourse.org/lectures/ but of course these are just scratching the surface and you can get as advanced as you want like https://mml-book.github.io/
Re:Where to learn the math for data science? by s1t5 on Wednesday July 29, 2020 @01:43PM (#23989534)
You need mainly probability, statistics, multivariable calculus and linear algebra. It's a lot of material to cover and it takes years. I would suggest to pick a mathematics course/book that's just above your current level (whatever that is) and start there.

That shouldn't stop you from doing kaggle competitions right now though if thats' what you like. Personally I dislike kaggle but there's no reason why you can't just grab a dataset, experiment with different algorithms and learn as you go.

Re:Where to learn the math for data science? by pizza on Wednesday July 29, 2020 @03:33PM (#23990699)
Google around and look for multiple lecturers’s powerpoint pdf slides online for classes named things like “intro to big data”, “beginning data analysis”, or something similar. I find that powerpoints are actually a decent way to learn math if the exposition and explanation of reasoning is half-decent. It’s good to have multiple examples to cross reference

You may have to learn a little multivariable calculus, or at the very least look at enough examples of its notation to be able to comfortably skim equations

Re:Where to learn the math for data science? by martinmartinez on Wednesday July 29, 2020 @05:52PM (#23991980)
Re:Where to learn the math for data science? by p1esk on Wednesday July 29, 2020 @09:16PM (#23993338)
Learn about chain rule. Then learn how to multiply two matrices. Look up the definition of a vector norm. This will cover almost everything you need to know.