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

61–70 of 106 posts

Re: What Data Can’t Do

#61

Post by wodenokoto »

polytely wrote: post #15
wodenokoto wrote: post #2

The article seems to stop pretty early, as if something is missing. It’s an anecdote about a government incentive to have doctors see patients within 48 hours causing doctors to refuse scheduling patients later than 48 hours in order to get the incentive bonus. This is not an example of limits of data, but an example of perverse incentives.

Did you use reader mode? because I noticed that when I use firefox's reader mode it will cut off part of the article on the New Yorker site.

On iOS. Initially used reader mode, but switched because it seemed cut off.

But also without reader mode I can’t see more than the NHS anecdote.

Re: What Data Can’t Do

#62

Post by jplr8922 »

Confusing performance metrics and strategical objective is not a data problem, it is a human problem. It happens to a lot of people outside the usual Blair-WhiteNationalist-IQ crowd. I do not think that advanced technical knowledge in ML or stats is required to avoid this mistake ; it is the ability to perform valid counterfactuals statements.

A good example of what I mean can be found on wikipedia :

His instinctive preference for offensive movement was typified by an answer Patton gave to war correspondents in a 1944 press conference. In response to a question on whether the Third Army's rapid offensive across France should be slowed to reduce the number of U.S. casualties, Patton replied, "Whenever you slow anything down, you waste human lives."[103]

https://en.wikipedia.org/wiki/George_S._Patton

Here, US general Patton is not confounding a performance metric (number of casualities) with strategic goal (winning the war). His counterfactual statement could be that ''if we slow things down, you are simply delaying future battles and increase the total number of casualties in order to achieve victory''.

I'm not suprised at Blair decision. When we choose leaders, do we favor long term strategic thinkers, or opportunistic pretty faces?

Re: What Data Can’t Do

#63

Post by _dps »

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…

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 affect your objective function) on average one time, and could very well not appear at all.

- can I bound how badly I will suffer if my system is allowed to operate in the wild on such an input?

DL gives no additional tools to engage these questions.

Re: What Data Can’t Do

#64

Post by potatoman22 »

fractionalhare wrote: post #58
kvathupo wrote: post #32

Earlier quoted context omitted.

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…

> 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. Okay, so that's about the same as classical statistics. You're just waiving the requirement to know what the distribution is. You are still assuming there exists a distribution and that it holds in the future when you apply the model…

Of course there's a distribution behind the data. The parent commenter was saying not all machine learning techniques need to know that distribution, as a refute to their parent comment.

Re: What Data Can’t Do

#65

Post by carlosf »

sfink wrote: post #50
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…

Personally, I think the main problem with ML is simpler: it works well for interpolation, and is crap for extrapolation. If the outputs you want are well within the bounds of your training data set, ML can do wonders. If they aren't, it'll tell you that in 20 years everyone will be having -0.2 children and all the other species on the planet will start having to birth human babies just so they can be thrown into the…

I agree, but that's equivalent to my original claim.

Being bad at extrapolation is a consequence of assuming all training data can describe your phenomena distribution and being wrong.

Re: What Data Can’t Do

#66

Post by fractionalhare »

potatoman22 wrote: post #64
fractionalhare wrote: post #58

Earlier quoted context omitted.

> 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. Okay, so that's about the same as classical statistics. You're just waiving the requirement to know what the distribution is. You are still assuming there exists a distribution and that it holds in the future when you apply the model…

Of course there's a distribution behind the data. The parent commenter was saying not all machine learning techniques need to know that distribution, as a refute to their parent comment.

I know what they're saying, I even reiterate it in my second sentence. My point is that doesn't protect you from the distribution changing, which is a problem that applies to machine learning and classical statistics.

This is in support of the GP comment: while you can loosen your assumptions about what the underlying distribution is and don't literally need to know it, you can't get away from the fundamental limitations of statistics. Which is the original topic we're talking about.

Re: What Data Can’t Do

#68

Post by temp8964 »

cloogshicer wrote: post #60
temp8964 wrote: post #59

This article is totally gibberish. It's a terrible mixture of many unrelated things. Just because those things all have something to do with data (anything can be presented in numeric form), it does not make their issues are about data. First, the Tony Blair example is not about data. It is a failure of government planning. It's wrong politics and wrong economy. The G.D.P. example is laughable. G.D.P. is never intend…

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.

Re: What Data Can’t Do

#69

Post by RobinL »

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”?

yeah, agreed - a good understanding of the model's statistical assumptions can often help you make the model more robust and also give you ideas for what types of feature engineering are likely to work.

Re: What Data Can’t Do

#70

Post by kvathupo »

_dps wrote: post #63
kvathupo wrote: post #32

Earlier quoted context omitted.

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…

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.
Post reply on HN