Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Machine Learning Flashcards

machinelearningflashcards.com

Post reply on HN
Print view View original

21–30 of 41 posts

Re: Machine Learning Flashcards

#21

Post by autokad »

uoaei wrote: post #19

This is encouraging exactly what I dislike about the current ML-in-industry space, namely the fetishism surrounding being able to describe the rote steps of an algorithm and all this "X from scratch" stuff. It's good to know "this algorithm is used for binary classification" but there are so many subtleties to how the data is reckoned with through these algorithms and how that particular representation of the problem…

> "VMs would be a terrible idea because the hinge loss only focuses on two data points and essentially ignores all the rest"

Not true, In my experience fitted SVMs have thousands of support vectors. The hinge loss is supposed to be less sensitive to outliers.

In general, I think SVMs are a 'terrible idea' because you can often get better fits at much faster run times with gradient boosting or you have to spend a lot of time getting the kernel just right.

Re: Machine Learning Flashcards

#22

Post by uoaei »

autokad wrote: post #21
uoaei wrote: post #19

This is encouraging exactly what I dislike about the current ML-in-industry space, namely the fetishism surrounding being able to describe the rote steps of an algorithm and all this "X from scratch" stuff. It's good to know "this algorithm is used for binary classification" but there are so many subtleties to how the data is reckoned with through these algorithms and how that particular representation of the problem…

> "VMs would be a terrible idea because the hinge loss only focuses on two data points and essentially ignores all the rest" Not true, In my experience fitted SVMs have thousands of support vectors. The hinge loss is supposed to be less sensitive to outliers. In general, I think SVMs are a 'terrible idea' because you can often get better fits at much faster run times with gradient boosting or you have to spend a lot…

I'm eschewing kernels entirely and just sticking to linear models for reasons around interpretability (need to convert the model's coefficients+intercept to an explicit Boolean statement). But you're right that kernel methods are more flexible (maybe too flexible).

Re: Machine Learning Flashcards

#25

Post by SpaceManNabs »

SpaceManNabs wrote: post #17

There are people saying here that is not worth it. I tend to agree but for different reasons. Flashcards with spaced repetition can help you learn mathy concepts, as Michael Niesel and others have shown. As high_derivative notes, to make the process worthwhile, you need to make the notes yourself in order to internalize chunks that are worthwhile to you. That means your own definitions that you are failing to remembe…

Been too long since I commented so I can't edit.

I think these are fine if you are trying to review concepts while on a train or something. Emphasis on REVIEW and revisiting only, not learning.

Re: Machine Learning Flashcards

#27

Post by notefuel »

bsanr wrote: post #7

Darn, got my hopes up. https://www.evernote.com/l/ABq8HaHnNtlBRJNb62FXRx-imGp-s0-jY...

We recently launched precisely this - a note-taking app that lets people add questions/identify salient terms for review as flashcards, and ML that applies those questions to auto-generate fill-in-the-blank for any written material.

Would love to hear any feedback you have. The app is called Notefuel: https://apps.apple.com/us/app/notefuel/id1458567718

Re: Machine Learning Flashcards

#28

Post by chrisalbon »

benrbray wrote: post #3
alexilliamson wrote: post #2

These come up a lot on data science twitter. I never understand who the target audience could possibly be.

Yeah, this seems like a terrible way to actually learn ML. Maybe it's for business-type people who want to stay up to date on the buzzwords?

totes agree

Re: Machine Learning Flashcards

#29

Post by Der_Einzige »

uoaei wrote: post #19

This is encouraging exactly what I dislike about the current ML-in-industry space, namely the fetishism surrounding being able to describe the rote steps of an algorithm and all this "X from scratch" stuff. It's good to know "this algorithm is used for binary classification" but there are so many subtleties to how the data is reckoned with through these algorithms and how that particular representation of the problem…

On the job experience about which ML algorithm is more useful for a task is certainly helpful.

But, knowing how to implement "X from scratch" gives you far more useful information about how and why a particular algorithm is suitable for a certain task. We should encourage people to reimplement algorithms from scratch (for fun, not production usage) so that they understand how the sausage is made.

Also, SVMs support a whole lot of different loss functions. I suspect that a properly tuned SVM will do better than a properly tuned logistic regression.

Post reply on HN