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

What Data Can’t Do

newyorker.com

Post reply on HN
Print view View original

71–80 of 106 posts

Re: What Data Can’t Do

#71

Post by currymj »

carlosf wrote: post #43
currymj wrote: post #28

Earlier quoted context omitted.

i think this actually gets at what makes applied ML distinct from statistics as a practice, even though there is a ton of overlap. statisticians make assumptions 1 and 2, and think of themselves as trying to find the "correct" parameters of their model. people doing applied ML typically assume they don't know 1 (although they might implicitly make some weak assumptions like sub-gaussian to avoid fat tails, etc.) and…

I disagree, every ML model has some implicit statistical assumption, which is often not well understood by practitioners. At minimum you must assume your underlying process is not fat tailed. If it is, then your training/validation/test data might never be enough to make reliable predictions and your model might break constantly in prod. BTW shifting distributions and fat tailed distributions are sort of equivalent,…

I don't disagree with any of that, but I still think a responsible, clear-thinking ML practitioner can avoid having to assume the form of the data-generating process, depending on their application.

In some cases if you care about PAC generalization bounds, it's even the case that the bounds do actually hold for all possible distributions.

Re: What Data Can’t Do

#72

Post by _dps »

kvathupo wrote: post #70
_dps wrote: post #63

Earlier quoted context omitted.

It is absolutely untrue that DL is immune to fat-fail problems, and it is important that no one operate mission critical systems under this assumption. The two fat tail questions one has to engage are: - is it possible that a catastrophic input might be lurking in the wild that would not be present in a typical training set? Even with a 1M instance training set, a one-in-a-million situation will only appear (and affe…

I don't quite follow: is not what you described a flaw fundamental to all forecasting; that is, the occurrence of a gross outlier? I should clarify that DL doesn't suffer from the same problem the normality condition has on fat-tails: a failure to capture the skew of the distribution.

It's not characteristic of all forecasting, only purely empirical forecasting.

Definitionally, the only way to reason about risk that doesn't appear in training data is non-empirical (e.g. a priori assumptions about distributions, or worst cases, or out-of-paradigm tools like refusing to provide predictions for highly non-central inputs).

DL is not any better (or worse) than any other purely empirical method at answering questions about fat-tail risk, and the only way to do better is to use non-empirical/a-priori tools. Obviously the tradeoff here is that your a priori assumptions can be wrong, and that too needs to be included in your risk model (see e.g. Robust Optimization / Robust Control).

Re: What Data Can’t Do

#73

Post by not2b »

dsaavy wrote: post #38
ppod wrote: post #34

This author has published a couple of articles like this at the New Yorker They all have this in common: the author works through some interesting and in some ways unusual cases where data or statistics have been improperly or naively applied, with some social costs. I really enjoy the articles themselves. Then the New Yorker packages it up with a cartoon and a headline and subheadline like "Big Data: When will it ea…

> *Numbers don’t lie, except when they do. Harford is right to say that statistics can be used to illuminate the world with clarity and precision. They can help remedy our human fallibilities. What’s easy to forget is that statistics can amplify these fallibilities, too. As Stone reminds us, “To count well, we need humility to know what can’t or shouldn’t be counted.”* I do have a problem with her conclusion here. Ar…

One of the points is that the act of collecting the numbers and making decisions based on them can change the underlying behavior. The numbers can be perfectly correct (how many cases does the IT department get? How long does it take on average to resolve the issue?). The goal can be correct (we want to get issues resolved faster). But as soon as you try to manage people based on those perfectly valid numbers, but things often happen.

Re: What Data Can’t Do

#74

Post by not2b »

temp8964 wrote: post #68
cloogshicer wrote: post #60

Earlier quoted context omitted.

Gotta disagree here. This article is acknowledging a pattern, that data is misused in many different areas. I think the problem goes even deeper, which is a misunderstanding of the scientific method. Good discussion about this topic here: https://news.ycombinator.com/item?id=26122712

"data is misused in many different areas" is not a valuable / informative point. There are many wrongs seem to have something to do with data, but in fact they are not. Like socialist economy planning will eventually fail, but then you would say they misused data. It seems relevant, but misusing the data is not the real cause of their failure at all.

Replace "socialist" with "large company". The companies gather data, establish metrics, and manage to those numbers, and often bad things result. Ever been in a company where some internal support function goes to hell because its top manager's bonus depends on a metric, and they can improve that metric by refusing to support the users (find excuses to close IT support calls without fixing the issue, etc).

Re: What Data Can’t Do

#77

Post by contravariant »

erichahn wrote: post #22
carlosf wrote: post #4

I am increasingly worried with people applying ML in everything without any rigour. Statical inference generally only works well in very specific conditions: 1 - You know the distribution of the phenomenon under study (or make an explicit assumption and assume the risk of being wrong) 2 - Using (1), you calculate how much data you need so you get an estimation error below x% Even though most ML models are essentially…

Isn't the point of ML exactly that you don't know the underlying distribution? How is this ever assumed in any way? ML is not parametric statistics.

Well, all optimization problems are equivalent to a maximum likelihood estimate for a corresponding probability distribution so you may make more implicit assumptions than you think.

Typical ML methods just have a huge distribution space that can fit almost anything from which they pick just 1 option. This has two downsides:

Since your distribution space is several times too large by design you lose the ability to say anything useful about the accuracy of your estimate, other than that it is not the only option by far.

Since you must pick 1 option from your parameter space you may miss slightly less likely explanations that may still have huge consequences, which means your models tend to end up overconfident.

Re: What Data Can’t Do

#78

Post by splithalf »

nickforr wrote: post #52
RobinL wrote: post #31

Earlier quoted context omitted.

Yes - this is pretty much exactly how I explain the difference between machine learning and statistics. Despite using similar models, the expertise required for 'doing statistics' (statistical inference) is actually very different from machine learning. Machine learning fits into the 'hacker mentality' well - try stuff out see what works. To do statistical inference effectively, you really do need to spend time learn…

But without some statistical knowledge, isn’t there a risk of a lack of understanding about the robustness of “what works”?

Statistical knowledge doesn’t remove that risk. The extent to which it even lowers the risk is a question that could be answered empirically.

Re: What Data Can’t Do

#79

Post by temp8964 »

not2b wrote: post #74
temp8964 wrote: post #68

Earlier quoted context omitted.

"data is misused in many different areas" is not a valuable / informative point. There are many wrongs seem to have something to do with data, but in fact they are not. Like socialist economy planning will eventually fail, but then you would say they misused data. It seems relevant, but misusing the data is not the real cause of their failure at all.

Replace "socialist" with "large company". The companies gather data, establish metrics, and manage to those numbers, and often bad things result. Ever been in a company where some internal support function goes to hell because its top manager's bonus depends on a metric, and they can improve that metric by refusing to support the users (find excuses to close IT support calls without fixing the issue, etc).

Yes. But then some would say the company fail because it "misused data", but it was not the real cause of the company's failure. Any project involves using data could blame the failure on "misused data" which is an useless conclusion.

Re: What Data Can’t Do

#80

Post by clircle »

kvathupo wrote: post #32
carlosf wrote: post #4

I am increasingly worried with people applying ML in everything without any rigour. Statical inference generally only works well in very specific conditions: 1 - You know the distribution of the phenomenon under study (or make an explicit assumption and assume the risk of being wrong) 2 - Using (1), you calculate how much data you need so you get an estimation error below x% Even though most ML models are essentially…

As currymj commented, this isn't accurate for ML, only for classical statistics. In ML (or more specifically deep learning), we make no distribution-based assumptions, other than the fundamental assumption that our training data is "distributed like" our test data. Thus, there aren't issues with fat-tailed distributions since we make no such normality assumptions. Indeed, with the use of autoencoders, we don't assume…

The only reason that this may not be accurate for ML is because machine learners generally make no attempt to quantify their uncertainty in their predictions with e.g. confidence intervals or prediction intervals.

And there is a whole field of non-parametric statistics that doesn't make distribution assumptions.

Post reply on HN