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

A Student's Guide to Preparing for Data Science Interviews

acheronanalytics.com

Post reply on HN
Print view View original

51–60 of 61 posts

Re: A Student's Guide to Preparing for Data Science Interviews

#51

Post by gaius »

achompas wrote: post #40
reader5000 wrote: post #24

Earlier quoted context omitted.

Technically anybody who uses Excel is a "data scientist". Just got to get the right buzzwords.

This is absurd and false. This person is an analyst of some sort. Maybe this holds in the consulting world? It definitely does not hold in the tech world, IME.

This is absurd and false. This person is an analyst of some sort.

And what is a data scientist then, if their work does not involve analysing data and presenting their analysis?

99.9% of "data science" is exactly what people used to do in tools like Excel, MATLAB, even SQL, just in Jupyter instead. On a Mac while sipping a latte.

Re: A Student's Guide to Preparing for Data Science Interviews

#52

Post by reader5000 »

achompas wrote: post #48
reader5000 wrote: post #42

Earlier quoted context omitted.

>I think it's more like not hiring a big name coding competition winner because they never bothered to learn how to use version control, or any coding best practice, or any language other than C. Depends on what you're hiring for, but I'll take "competition winner with no version control" over "average programmer with expert VC capabilities". >Bayes rule isn't some kind of deep magic Yes, it's largely conceptually ob…

> Yes, it's largely conceptually obsolete. I'm sorry, what? How did you arrive at a point where you believe this is true? This is like calling compilers "obsolete." Is it because you believe deep learning has "taken over" or something?

Try to derive e.g. a face detector from bayes theorem. You immediately arrive at computationally intractable sums/integrals. Yet, we have super-human image classifiers. Therefore, bayes theorem is obsolete. Sure, you can try to retrofit bayes theorem on top of a neural net, but who cares?

Re: A Student's Guide to Preparing for Data Science Interviews

#53

Post by achompas »

reader5000 wrote: post #52
achompas wrote: post #48

Earlier quoted context omitted.

> Yes, it's largely conceptually obsolete. I'm sorry, what? How did you arrive at a point where you believe this is true? This is like calling compilers "obsolete." Is it because you believe deep learning has "taken over" or something?

Try to derive e.g. a face detector from bayes theorem. You immediately arrive at computationally intractable sums/integrals. Yet, we have super-human image classifiers. Therefore, bayes theorem is obsolete. Sure, you can try to retrofit bayes theorem on top of a neural net, but who cares?

> You immediately arrive at computationally intractable sums/integrals.

So we instead sample from that posterior.

Unless you think MCMC is also obsolete, in which case I’ll see myself out.

Re: A Student's Guide to Preparing for Data Science Interviews

#54

Post by achompas »

gaius wrote: post #51
achompas wrote: post #40

Earlier quoted context omitted.

This is absurd and false. This person is an analyst of some sort. Maybe this holds in the consulting world? It definitely does not hold in the tech world, IME.

This is absurd and false. This person is an analyst of some sort. And what is a data scientist then, if their work does not involve analysing data and presenting their analysis? 99.9% of "data science" is exactly what people used to do in tools like Excel, MATLAB, even SQL, just in Jupyter instead. On a Mac while sipping a latte.

> And what is a data scientist then, if their work does not involve analysing data and presenting their analysis?

This is a dead giveaway that you have no idea what you’re talking about. You’ve captured about 5% of my work.

The rest of the time, I’m writing software (ETL pipelines or real-time services, including tests), debugging some distributed system, collecting or cleaning data, or gathering requirements and developing feature specs with other folks.

Fortunately for you, you nailed the Mac-using latte-drinking part!

EDIT: Reading your comment history on regression and k-means. you _do_ know what you’re talking about. It is hard to get models into production, so I’m surprised to see your snark here. What gives? Do you have experience with DS who don’t deliver?

Re: A Student's Guide to Preparing for Data Science Interviews

#55

Post by gaius »

achompas wrote: post #54
gaius wrote: post #51

Earlier quoted context omitted.

This is absurd and false. This person is an analyst of some sort. And what is a data scientist then, if their work does not involve analysing data and presenting their analysis? 99.9% of "data science" is exactly what people used to do in tools like Excel, MATLAB, even SQL, just in Jupyter instead. On a Mac while sipping a latte.

> And what is a data scientist then, if their work does not involve analysing data and presenting their analysis? This is a dead giveaway that you have no idea what you’re talking about. You’ve captured about 5% of my work. The rest of the time, I’m writing software (ETL pipelines or real-time services, including tests), debugging some distributed system, collecting or cleaning data, or gathering requirements and dev…

What gives? Do you have experience with DS who don’t deliver?

I have experience of DS who define what they do by the tools they use, not the results they deliver, it's a pet peeve of mine :-)

Thanks for going back and making the edit!

Re: A Student's Guide to Preparing for Data Science Interviews

#56

Post by reader5000 »

achompas wrote: post #53
reader5000 wrote: post #52

Earlier quoted context omitted.

Try to derive e.g. a face detector from bayes theorem. You immediately arrive at computationally intractable sums/integrals. Yet, we have super-human image classifiers. Therefore, bayes theorem is obsolete. Sure, you can try to retrofit bayes theorem on top of a neural net, but who cares?

> You immediately arrive at computationally intractable sums/integrals. So we instead sample from that posterior. Unless you think MCMC is also obsolete, in which case I’ll see myself out.

You're right, but a) you have comp efficiency issues with MCMC, and b) just empirically MCMC models don't work as well as gradient descent + NN for many tasks.

Re: A Student's Guide to Preparing for Data Science Interviews

#57

Post by achompas »

reader5000 wrote: post #56
achompas wrote: post #53

Earlier quoted context omitted.

> You immediately arrive at computationally intractable sums/integrals. So we instead sample from that posterior. Unless you think MCMC is also obsolete, in which case I’ll see myself out.

You're right, but a) you have comp efficiency issues with MCMC, and b) just empirically MCMC models don't work as well as gradient descent + NN for many tasks.

And you don't have computational efficiency issues with NNs?

We're also ignoring the benefits of a posterior distribution, which is useful for understanding the data-generating process.

Re: A Student's Guide to Preparing for Data Science Interviews

#58

Post by reader5000 »

achompas wrote: post #57
reader5000 wrote: post #56

Earlier quoted context omitted.

You're right, but a) you have comp efficiency issues with MCMC, and b) just empirically MCMC models don't work as well as gradient descent + NN for many tasks.

And you don't have computational efficiency issues with NNs? We're also ignoring the benefits of a posterior distribution, which is useful for understanding the data-generating process.

Yeah of course. I can't explain to you why NNs outperform bayesian approaches, probably just NNs are capturing the correct type of prior for vision/language tasks. And yeah bayesian models are more interpretable but when you have millions of latent variables I'm not sure interpretability is a thing.

Re: A Student's Guide to Preparing for Data Science Interviews

#59

Post by achompas »

reader5000 wrote: post #58
achompas wrote: post #57

Earlier quoted context omitted.

And you don't have computational efficiency issues with NNs? We're also ignoring the benefits of a posterior distribution, which is useful for understanding the data-generating process.

Yeah of course. I can't explain to you why NNs outperform bayesian approaches, probably just NNs are capturing the correct type of prior for vision/language tasks. And yeah bayesian models are more interpretable but when you have millions of latent variables I'm not sure interpretability is a thing.

Yep, we arrived at my larger point: if you care about interpretability, NNs are horrible and Bayesian techniques are pretty damn great.

Re: A Student's Guide to Preparing for Data Science Interviews

#60

Post by reader5000 »

achompas wrote: post #59
reader5000 wrote: post #58

Earlier quoted context omitted.

Yeah of course. I can't explain to you why NNs outperform bayesian approaches, probably just NNs are capturing the correct type of prior for vision/language tasks. And yeah bayesian models are more interpretable but when you have millions of latent variables I'm not sure interpretability is a thing.

Yep, we arrived at my larger point: if you care about interpretability, NNs are horrible and Bayesian techniques are pretty damn great.

Well certainly, but interpretability is obsolete.
Post reply on HN