14 ms·

> Nobody knew or even cared what the difference was between good and bad data science work. Meaning you could absolutely suck at your job or be incredible at it

> Nobody knew or even cared what the difference was between good and bad data science work. Meaning you could absolutely suck at your job or be incredible at it and you’d get nearly the same regards in either case. In my experience it's even a little bit worse than that. Approaches that are wrong from a statistics point of view are more likely to generate impressive seeming results. But the flaws are often subtle. A common one I've seen quite many times is people using a flawed validation strategy (e.g. one which rewards the model for using data "leaked" from the future), or to rely on in-sample results too much in other ways. Because these issues are subtle, management will often not pick up on them or not be aware that this kind of thing can go wrong. With a short-term focus they also won't really care, because they can still put these results in marketing materials and impress most outsiders as well.

CapmCrackaWaka 4y ago
I've seen this a LOT in my professional group. Many people (who often have PhDs!!) I interview for data science positions seem to know absolutely nothing about the algorithms they use professionally, or how to optimize them, or why they are a good fit for their use case, etc etc etc. I usually see through LinkedIn that these same people are now in impressive-sounding positions at other companies. I had one candidate who was in charge of a multi-armed-bandit project at their current company. I asked them how it worked, and how they settled on that. Their response was "you know, I'm not really sure, the code was set up when I got there". He had been there for over a year, and could tell me nothing! > A common one I've seen quite many times is people using a flawed validation strategy (e.g. one which rewards the model for using data "leaked" from the future), or to rely on in-sample results too much in other ways. It's funny you mention this, we have a direct competitor who does this and advertises flawed metrics to clients. Often times our clients will come back to us saying "XYZ says they can get better performance", the performance in this case being something which is simply impossible without data leakage or some flawed validation strategy.
- adamsmith143 4y ago
  Where are these jobs where you can interview this badly and still get hired because in my experience DS interviews are extremely hard and often expect people to have very high Stats skills as well as Data Structures/Algo skills at FAANG level.
  - semi-extrinsic 4y ago
    These days if you have a company selling cat food or rivets for aerospace or providing taxi swrvice to a random city, or whatever, they might have a few data scientists helping them make "optimized" business choices. Obviously they won't have a very adcanced recruiting process for that.
    - quickthrower2 4y ago
      Like the market for lemons: https://en.m.wikipedia.org/wiki/The_Market_for_Lemons https://en.m.wikipedia.org/wiki/The_Market_for_Lemons
  - noelsusman 4y ago
    It's different at a lot of non-tech companies. I'm in the nonprofit world and my interview barely had any technical component at all.
    - bo1024 4y ago
      To be able to tell whether a candidate is good, the hiring team has to be expert! No chicken, no egg.
  - bootsmann 4y ago
    I think the issue here is that "data science" encompasses two very distinct branches of work. One answers to business needs and the other produces data based solutions for the product itself i.e you might have a data scientist who A/B tests your website design so you minimize your churn rate and the other is the team at uber eats who maintains the recommendation engine. While the distinction might not always be as sharp, the former makes up the bulk of data scientists in the market (and I suspect the OP is in that boat) with comparably simple interviews while the rest is the 5 step interview process with hackerrank test you are more familiar with.
- Radim 4y ago
  > clients will come back to us saying "XYZ says they can get better performance" Oh yes, good old marketing. Along with buying off "Industry Awards" – hey, we're objectively the "Best cybersecurity company of 2022!" With a matching "platinum/gold badge" to go on our website! Or buying a place in the "10 Best Products for X" and "Independent X-vs-Y Comparison", another classic. Because it works. Are your customers not sophisticated? Are they unable (or unwilling) to follow up on defects and outright lies? Or reality simply doesn't matter all that much to them? Humans LOVE a good story more than reality, after all. Then your contribution as an engineer to your company's success, and hence its longevity and your job security, is strictly inferior to that of marketing. Not everything is the work of evil marketers – a lot of the supplied BS is in response to an existing demand for BS.
  - CapmCrackaWaka 4y ago
    > Are your customers not sophisticated? Are they unable (or unwilling) to follow up on defects and outright lies? You would probably be depressed if you knew who our customers were, and how technologically unsophisticated they are.
    - Foobar8568 4y ago
      I manage at a client an application which is the actual leader (most top right and by far) in Gartner magic quadrant for its category, and for years, I have never seen a product this bad, where the implementors and supports are clueless of their own product. And obviously it's buggy as hell. Lies and deceptions.
      - jacobyoder 4y ago
        The people who make the decisions don't use the product. That's almost always the root cause of this stuff. I worked on a system for my state - another vendor came in and 'took over' all the functionality my system handled. Supposedly. 7 years later, my system powers the exception to the mandate to 'use system X', because... they refuse to provide the functionality that they sold the state. Contractually, "we provide feature ABC", but the reality is.. they don't. I even provided them our code to use - it was paid for with public money, they should just integrate it and then sell it to other people to make their product better. They can't even be bothered to take the code and integrate it... they prefer to continually lie and say "we provide feature ABC" when... they don't. It's beyond insane. A large majority of the people on the ground know it's bad/lacking/broken, but ... they have 0 voice in the matter.
- f1shy 4y ago
  Is this the US? I'm concern about the extremely low level or the bar to get a PhD in Europe... and I'm wondering if that is a global problem, or only Europe.
- DenisM 4y ago
  > XYZ says they can get better performance Can you do your analysis both ways? Give your customers both, then tell them you method is more modern, but if they want outdated methods you have those too.
6gvONxR4sf7o 4y ago
This is the defining pain point for data science, in my experience. There’s no simple ground truth to test competence against. If someone tells you that the data says their work is good, the only real way to know if they’re right or wrong is to look at what the data says yourself. If 99% of the work is building and 1% is checking something like latency, then you’re likely to have more than one set of eyeballs on that 1%. But if 99% of the work is putting the data together and doing the analysis, then you’re unlikely to have more than one person ever look at that part. So incompetence goes unchecked (or worse, it is rewarded).
- vsareto 4y ago
  That's the same for many tech jobs. Competence is often only a local thing, subject to politics, reputation, and appearances. There's also no ground truth because the ground changes so fast. No one knows if the technologies mentioned in the OP will be popular 5-10 years from now.
  - jimbokun 4y ago
    But if the page loads slowly or the UI is unresponsive, people notice. The output of Data Science is harder for non-specialists to evaluate.
    - vsareto 4y ago
      For establishing competence, you still have to dig in to see what caused the slowness. A regular user can't tell you that.
      - pessimizer 4y ago
        > For establishing competence, you still have to dig in to see what caused the slowness. Not as management. You just have to see that other people's similar sites are not slow with the same resources, therefore it is possible for your site not to be slow. You don't have to know why you're failing to know that the totality of the people you hired were not as good as the people those others hired. This is of course barring management failure; but if you're failing at management, that's about the same as saying that your engineers were under-resourced. Engineering competence is largely composed of the skills to figure out what is causing problems e.g. slowness. If you can't figure out what is causing the slowness, your engineers aren't good enough to figure out what is causing the slowness, qed. That's different than data science.
chasely 4y ago
I've been pitched by many "data-driven" vendors offering predictions. They often have very impressive accuracy metrics (RMSE, R2, etc). When I dive into the details these metrics are often reported using in-sample predictions. I see this pointing to any of the following: a) DS teams overpromising the accuracy of their approaches b) marketing driving the narrative and DS getting pulled along c) incompetence from the DS team
- alexpetralia 4y ago
  Are these inferential statistics not designed to be in-sample? I would imagine predictive statistics use more out-of-sample metrics like precision and recall.
  - antipaul 4y ago
    If you do it right… That’s the problem: these metrics often come from overfitted or in-sample data, and are completely unrealistic when it comes to expected generalization performance. I’m at the point where I never trust performance metrics anymore. Or rather, the worse they are, the more I trust them!
  - ZephyrBlu 4y ago
    I feel like you might be conflating a couple of things, though I'm not a DS so could be off base here. My reading of the OP's description is that the vendors were offering interpolative predictions, but did not use a test/train split of data. This is in contrast to extrapolative predictions which I would call out-of-sample. Thus due to not using a test/train split, they achieved extremely good accuracy because they were testing on the same data they trained on. Even though this is "in-sample", you can't use the same data for testing and training.
whatever1 4y ago
Blame statistics for that. Wrong outcome? Well you were unlucky you fell into the 1% error range. Correct outcome? You totally predicted it correctly. There is literally no way you can screw up something in statistics and not being able to make up a story to defend your approach.
- lagt_t 4y ago
  You don't look at single outcomes with statistics.
  - whatever1 4y ago
    See? “Better luck next time”. Not being mean to you, just showing how typically the goal posts are moved. To give you an example from physics, if you find just one experiment that goes against your model, you immediately invalidate the model. You don’t just make grand claims that the model in general works.
    - cjglo 4y ago
      Could be wrong here, but in physics and most natural sciences, you don’t throw away your model if you have one experiment against it. Usually isn’t it looking for an experiment that proves it and is repeatable? If I discover a new element in one experiment, the results are published. After publication, many labs will try to repeat and its not taken away if one can’t do it. Only if all can’t and it casts doubt on whether I did it in the first place.
      - Jensson 4y ago
        Scientific method? Models are disproved, not proved. Do data scientists not know about science? People usually understand how science works here on HN, but not in this thread. Example of a test that invalidated our old theory of gravity and validated Einsteins claims: https://en.wikipedia.org/wiki/Eddington_experiment https://en.wikipedia.org/wiki/Eddington_experiment This is how science is done. But apparently not data science.
etempleton 4y ago
I become wary any time someone utters the phrase, "show me the data" or any variation there of. There is a specific type of leader who thinks that within the data lurks a magical solution just waiting to be discovered. There is also the leader who uses data as a trump card to win arguments and these folks are perhaps even worse. This is not new. The origination of the phrase, "lies, damned lies, and statistics," can be traced to the 1800s. I propose the following update: There are three kinds of lies: Lies, damned lies, and data I am being glib, I of course do not think all data is inconsequential, rather it is more often used from a place of ignorance or a place of ill intent it is rendered, on the whole, useless.
- lotsofpulp 4y ago
  I have only heard “show me the data” when someone wants someone else to support a claim. I do not see why this would necessarily be a bad thing.
  - fn-mote 4y ago
    It's BS because the people asking for the data do not have the sophistication to actually do a reasonable _analysis_ of the data. Or criticize an existing analysis. Unfortunately, as many posters here are pointing out, there's plenty of ways to do a correct-looking analysis of the data to get evidence to support your agenda. Maybe your agenda is right and maybe it's not, but I'd love to hear a story of someone standing up and saying "your consultant submitted a report with glaring flaws, they should not be paid and you should reconsider X." It's more likely the little company just goes out of business or the big company buries the failure.
    - ifyoubuildit 4y ago
      So whats the alternative? "Just trust me"?
      - yunwal 4y ago
        The alternative is to trust in most circumstances (you did hire me after all), and do thorough analysis once in a while as a gut check.
_jx7j 4y ago
>Meaning you could absolutely suck at your job or be incredible at it and you’d get nearly the same regards in either case. One of the things I don't like about statements like this said in a Data Science context, is that they are true outside of Data Science as well. Executives make big decisions, managers make smaller decisions, nobody can evaluate how good/bad they really were for months or years. Engineers build something amazing, or build a house of cards, nobody cares as long as the money people are happy, even if the business use case turns out to be wrong in the long run. >With a short-term focus they also won't really care, because they can still put these results in marketing materials and impress most outsiders as well. Forget Data Science, you see this in KPIs as well. Say a crappy metric has to be moved by Q2 next year and people will destroy the company to move it. I feel like Data Science is just one of those areas where you are exposed to a wider range of people and get to feel the full crapola of the insanity of working in a corporation. For lots of roles (e.g. Engineering) you get to hide in a hole behind layers of people and not see some of this insanity.
- scottLobster 4y ago
  Not to get too off topic, but as a 35 year old engineer it seems the world in general has far fewer consequences than I was raised to expect. Everything from businesses with bullshit ideas flourishing at a loss, to January 6 even being possible (politics aside I expected the Capitol Police to crack a lot more skulls than they did once people started smashing windows), to the whole FTX situation and the tepid response in the media/government, to petty crime being outright tolerated, to in my own career I've at times burned through enough money badly enough (albeit with good intentions) that I thought I was going to be fired, only to be told in a performance review I was doing a good job (grateful to stay employed but WTF, I would have fired or at least demoted me). Importantly, the motivation for this lack of consequence doesn't seem to stem from a desire for forgiveness or positive reinforcement or any mechanism that might make things better. It seems like there's a general apathy/nihilism that's growing in society, whereas by contrast my entire education from childhood up I was held to strict standards and reliably punished when I failed to meet them, and this was in US public schools (albeit a highly ranked school district) and a public university. That or I was just raised in a bubble, and the historical examples I referenced growing up and reference to this day are just a case of survivorship bias, and all the bullshit that was alongside them back in the day has simply been forgotten. I'm not sure, but it is disappointing how little people at large seem to give a shit. Maybe it's a side-effect of the obesity epidemic and people just have less energy or something
2devnull 4y ago
This gave me a chuckle. If you read the feature article you understand that this is also because management wants “decision driven data.” They have an idea and use ds to provide charts and tables to support their idea. The harder the idea is to support, the greater value data science is able to provide. I guess data science is inferior to research in this way. People care about research methods, rigor, etc… Maybe data scientists should adopt stricter standards, like actual scientists.
- samvher 4y ago
  I did read the article - some of the problems with judgements of work quality also come up with (hypothetical) well-intentioned truth-seeking non-political long-term-optimizing managers who just don't happen to be stats experts.
  - 2devnull 4y ago
    Sorry, wasn’t trying to imply you didn’t and I fully agree. Even managers that know stats can be busy or but into hype about ml or other shiny new things that they don’t have time or resources to deconstruct. This is another big problem with data science, “black box” systems and cargo cults. It’s easy to think “LLMs will change the world! We should use them, the competition will.”
dqpb 4y ago
> Approaches that are wrong from a statistics point of view are more likely to generate impressive seeming results. This is to be expected from an information theory point of view. It's why "fake news" will always be a thing.
jmount 4y ago
Super point. I can't resist repeating it back. When incorrect work outperforms correct work in superficial evaluation, it is then selected for.
mmcnl 4y ago
This is somewhat captured in the article as well. "Managers will say they want to make data-driven decisions, but they really want decision-driven data. If you strayed from this role– e.g. by warning people not to pursue stupid ideas– your reward was their disdain, then they’d do it anyway, then it wouldn’t work (what a shocker). The only way to win is to become a stooge." In science, a good scientific result can be bad for business. There is often little appreciation for the "science" in data science.
- throwingit0 4y ago
  >There is often little appreciation for the "science" in data science. It feels like even Google falls prey to this at times: they keep redoing the same A/B test until it comes up in favor of the change (or the designer whose pet project it is runs out of political capital, presumably).
whiplash451 4y ago
You should pay the price for data leakage very quickly in production. Does management look at slides or AB test dashboards?
- disgruntledphd2 4y ago
  People don't pay attention to production metrics, and a noisy problem (like marketing or whatnot) can often be pretty bad for a looooonnnnnggg time before anyone notices.
jrochkind1 4y ago
> Approaches that are wrong from a statistics point of view When OP talked about "the main bottleneck to my work" in terms of areas he would need to learn more about -- I was expecting him to talk about facility with statistical methods and using them appropriately! I'm not sure what to take from the fact that he never did! I would like to ask him what he thinks about that!
mumblemumble 4y ago
The problem is that nobody actually wants data science. They want data pseudoscience. And for the same reason that people tend to want pseudoscience instead of science in any other domain, too. Science is slow, tentative, and messy, and usually responds to questions with even more questions rather than with answers. Pseudoscience tends to be much more concerned with exuding confidence and providing clean-cut answers. It's what happens when a desire for science meets a need for instant gratification. Along the way, things like blinding and controls and watching for bias and validating assumptions tend to get dropped when they're inconvenient or difficult to explain. And they're always inconvenient and difficult to explain.
- onlyrealcuzzo 4y ago
  > The problem is that nobody actually wants data science. They want data pseudoscience. Technically, I think investors & owners would want the company to use real data science to improve products & maximize profits. Everybody in the middle just wants to use data to lie to get promoted faster - because you don't get promoted for actually doing a good job - you get promoted for convincing people you did a good job, and lying is a VERY useful / effective tool.
  - deckard1 4y ago
    Welcome to the world of OKRs/KPIs/Scrum, where everything's made up and the points don't matter.
  - NeutralCrane 4y ago
    > I think investors & owners would want the company to use real data science to improve products & maximize profits. This is based on the assumption that companies are focused on long term profits and stability, and I’m not sure why anyone believes that to be the case anymore. The vast majority of companies are run based on next quarter’s stock price or growth metrics. I worked on a newly formed data science team coming out of grad school that was tasked with taking some predictive initiatives that the company had relied on external consultants to produce, and implementing them in-house. The external team’s results always looked exactly like what the business wanted to hear, but they rarely played out in practice. This was in part because the underlying data quality was terrible, and the company wasn’t executing in a way that allowed anyone to actually answer the questions being asked. The consultants would just torture the data until they could come up with a report that would ensure the company would come back the following year. So we spent a lot of time trying pouring cold water into the business groups who saw data science as a magic wand that would conjure up more money at no cost. But we never were able to convince them to invest in anything that would take longer than a year. Anything that would require a change in their marketing or strategy executions that wouldn’t immediately deliver increased results was just a non-starter. But actual data science requires that kind of investment for long-term layoffs. So the data science team became figure-heads, never given the buy-in to actually make impact on business, but kept around so teams and leaders could tout being “data-driven” and throw “AI” and “machine-learning” into PR and marketing materials. You aren’t wrong about middle management is looking to get promoted faster. But every single individual from the employee looking for a promotion to the executive suite to the investors are addicted to incentive windows no longer than 6-12 months.
rgrieselhuber 4y ago
I’ve always disliked how data science was positioned within companies as well, it’s outside the critical path of product and engineering, which means it becomes a mere abstraction to management (e.g. “throw that problem to the data science team and see what they come up with”), resulting in very vague and abstract requirements and, hence, deliverables. I think there is huge value in the discipline and technologies, but it unfairly gets relegated when not integrated to the whole product / engineering process. Hence, the title / concept of Data Engineer seems like a much better fit for this role within many companies.
- rgavuliak 4y ago
  Yeah, as a Data Science manager I've experienced this pain a lot (not part of the critical path). I am now an Engineering Manager that works with a cross-functional team including FE/BE/DS/Devops and it's the most power I ever had to put Data Science in front of our clients in a meaningful way.
hcks 4y ago
> Because these issues are subtle, management will often not pick up on them or not be aware that this kind of thing can go wrong. Management is not your teacher at school, it is not there to check up your results make sense. Management mostly assumes you’re competent at your job.

CapmCrackaWaka 4y ago

I've seen this a LOT in my professional group. Many people (who often have PhDs!!) I interview for data science positions seem to know absolutely nothing about the algorithms they use professionally, or how to optimize them, or why they are a good fit for their use case, etc etc etc. I usually see through LinkedIn that these same people are now in impressive-sounding positions at other companies. I had one candidate who was in charge of a multi-armed-bandit project at their current company. I asked them how it worked, and how they settled on that. Their response was "you know, I'm not really sure, the code was set up when I got there". He had been there for over a year, and could tell me nothing! > A common one I've seen quite many times is people using a flawed validation strategy (e.g. one which rewards the model for using data "leaked" from the future), or to rely on in-sample results too much in other ways. It's funny you mention this, we have a direct competitor who does this and advertises flawed metrics to clients. Often times our clients will come back to us saying "XYZ says they can get better performance", the performance in this case being something which is simply impossible without data leakage or some flawed validation strategy.

adamsmith143 4y ago
Where are these jobs where you can interview this badly and still get hired because in my experience DS interviews are extremely hard and often expect people to have very high Stats skills as well as Data Structures/Algo skills at FAANG level.
- semi-extrinsic 4y ago
  These days if you have a company selling cat food or rivets for aerospace or providing taxi swrvice to a random city, or whatever, they might have a few data scientists helping them make "optimized" business choices. Obviously they won't have a very adcanced recruiting process for that.
  - quickthrower2 4y ago
    Like the market for lemons: https://en.m.wikipedia.org/wiki/The_Market_for_Lemons https://en.m.wikipedia.org/wiki/The_Market_for_Lemons
- noelsusman 4y ago
  It's different at a lot of non-tech companies. I'm in the nonprofit world and my interview barely had any technical component at all.
  - bo1024 4y ago
    To be able to tell whether a candidate is good, the hiring team has to be expert! No chicken, no egg.
- bootsmann 4y ago
  I think the issue here is that "data science" encompasses two very distinct branches of work. One answers to business needs and the other produces data based solutions for the product itself i.e you might have a data scientist who A/B tests your website design so you minimize your churn rate and the other is the team at uber eats who maintains the recommendation engine. While the distinction might not always be as sharp, the former makes up the bulk of data scientists in the market (and I suspect the OP is in that boat) with comparably simple interviews while the rest is the 5 step interview process with hackerrank test you are more familiar with.
Radim 4y ago
> clients will come back to us saying "XYZ says they can get better performance" Oh yes, good old marketing. Along with buying off "Industry Awards" – hey, we're objectively the "Best cybersecurity company of 2022!" With a matching "platinum/gold badge" to go on our website! Or buying a place in the "10 Best Products for X" and "Independent X-vs-Y Comparison", another classic. Because it works. Are your customers not sophisticated? Are they unable (or unwilling) to follow up on defects and outright lies? Or reality simply doesn't matter all that much to them? Humans LOVE a good story more than reality, after all. Then your contribution as an engineer to your company's success, and hence its longevity and your job security, is strictly inferior to that of marketing. Not everything is the work of evil marketers – a lot of the supplied BS is in response to an existing demand for BS.
- CapmCrackaWaka 4y ago
  > Are your customers not sophisticated? Are they unable (or unwilling) to follow up on defects and outright lies? You would probably be depressed if you knew who our customers were, and how technologically unsophisticated they are.
  - Foobar8568 4y ago
    I manage at a client an application which is the actual leader (most top right and by far) in Gartner magic quadrant for its category, and for years, I have never seen a product this bad, where the implementors and supports are clueless of their own product. And obviously it's buggy as hell. Lies and deceptions.
    - jacobyoder 4y ago
      The people who make the decisions don't use the product. That's almost always the root cause of this stuff. I worked on a system for my state - another vendor came in and 'took over' all the functionality my system handled. Supposedly. 7 years later, my system powers the exception to the mandate to 'use system X', because... they refuse to provide the functionality that they sold the state. Contractually, "we provide feature ABC", but the reality is.. they don't. I even provided them our code to use - it was paid for with public money, they should just integrate it and then sell it to other people to make their product better. They can't even be bothered to take the code and integrate it... they prefer to continually lie and say "we provide feature ABC" when... they don't. It's beyond insane. A large majority of the people on the ground know it's bad/lacking/broken, but ... they have 0 voice in the matter.
f1shy 4y ago
Is this the US? I'm concern about the extremely low level or the bar to get a PhD in Europe... and I'm wondering if that is a global problem, or only Europe.
DenisM 4y ago
> XYZ says they can get better performance Can you do your analysis both ways? Give your customers both, then tell them you method is more modern, but if they want outdated methods you have those too.

6gvONxR4sf7o 4y ago

This is the defining pain point for data science, in my experience. There’s no simple ground truth to test competence against. If someone tells you that the data says their work is good, the only real way to know if they’re right or wrong is to look at what the data says yourself. If 99% of the work is building and 1% is checking something like latency, then you’re likely to have more than one set of eyeballs on that 1%. But if 99% of the work is putting the data together and doing the analysis, then you’re unlikely to have more than one person ever look at that part. So incompetence goes unchecked (or worse, it is rewarded).

vsareto 4y ago
That's the same for many tech jobs. Competence is often only a local thing, subject to politics, reputation, and appearances. There's also no ground truth because the ground changes so fast. No one knows if the technologies mentioned in the OP will be popular 5-10 years from now.
- jimbokun 4y ago
  But if the page loads slowly or the UI is unresponsive, people notice. The output of Data Science is harder for non-specialists to evaluate.
  - vsareto 4y ago
    For establishing competence, you still have to dig in to see what caused the slowness. A regular user can't tell you that.
    - pessimizer 4y ago
      > For establishing competence, you still have to dig in to see what caused the slowness. Not as management. You just have to see that other people's similar sites are not slow with the same resources, therefore it is possible for your site not to be slow. You don't have to know why you're failing to know that the totality of the people you hired were not as good as the people those others hired. This is of course barring management failure; but if you're failing at management, that's about the same as saying that your engineers were under-resourced. Engineering competence is largely composed of the skills to figure out what is causing problems e.g. slowness. If you can't figure out what is causing the slowness, your engineers aren't good enough to figure out what is causing the slowness, qed. That's different than data science.

chasely 4y ago

I've been pitched by many "data-driven" vendors offering predictions. They often have very impressive accuracy metrics (RMSE, R2, etc). When I dive into the details these metrics are often reported using in-sample predictions. I see this pointing to any of the following: a) DS teams overpromising the accuracy of their approaches b) marketing driving the narrative and DS getting pulled along c) incompetence from the DS team

alexpetralia 4y ago
Are these inferential statistics not designed to be in-sample? I would imagine predictive statistics use more out-of-sample metrics like precision and recall.
- antipaul 4y ago
  If you do it right… That’s the problem: these metrics often come from overfitted or in-sample data, and are completely unrealistic when it comes to expected generalization performance. I’m at the point where I never trust performance metrics anymore. Or rather, the worse they are, the more I trust them!
- ZephyrBlu 4y ago
  I feel like you might be conflating a couple of things, though I'm not a DS so could be off base here. My reading of the OP's description is that the vendors were offering interpolative predictions, but did not use a test/train split of data. This is in contrast to extrapolative predictions which I would call out-of-sample. Thus due to not using a test/train split, they achieved extremely good accuracy because they were testing on the same data they trained on. Even though this is "in-sample", you can't use the same data for testing and training.

whatever1 4y ago

Blame statistics for that. Wrong outcome? Well you were unlucky you fell into the 1% error range. Correct outcome? You totally predicted it correctly. There is literally no way you can screw up something in statistics and not being able to make up a story to defend your approach.

lagt_t 4y ago
You don't look at single outcomes with statistics.
- whatever1 4y ago
  See? “Better luck next time”. Not being mean to you, just showing how typically the goal posts are moved. To give you an example from physics, if you find just one experiment that goes against your model, you immediately invalidate the model. You don’t just make grand claims that the model in general works.
  - cjglo 4y ago
    Could be wrong here, but in physics and most natural sciences, you don’t throw away your model if you have one experiment against it. Usually isn’t it looking for an experiment that proves it and is repeatable? If I discover a new element in one experiment, the results are published. After publication, many labs will try to repeat and its not taken away if one can’t do it. Only if all can’t and it casts doubt on whether I did it in the first place.
    - Jensson 4y ago
      Scientific method? Models are disproved, not proved. Do data scientists not know about science? People usually understand how science works here on HN, but not in this thread. Example of a test that invalidated our old theory of gravity and validated Einsteins claims: https://en.wikipedia.org/wiki/Eddington_experiment https://en.wikipedia.org/wiki/Eddington_experiment This is how science is done. But apparently not data science.

etempleton 4y ago

I become wary any time someone utters the phrase, "show me the data" or any variation there of. There is a specific type of leader who thinks that within the data lurks a magical solution just waiting to be discovered. There is also the leader who uses data as a trump card to win arguments and these folks are perhaps even worse. This is not new. The origination of the phrase, "lies, damned lies, and statistics," can be traced to the 1800s. I propose the following update: There are three kinds of lies: Lies, damned lies, and data I am being glib, I of course do not think all data is inconsequential, rather it is more often used from a place of ignorance or a place of ill intent it is rendered, on the whole, useless.

lotsofpulp 4y ago
I have only heard “show me the data” when someone wants someone else to support a claim. I do not see why this would necessarily be a bad thing.
- fn-mote 4y ago
  It's BS because the people asking for the data do not have the sophistication to actually do a reasonable _analysis_ of the data. Or criticize an existing analysis. Unfortunately, as many posters here are pointing out, there's plenty of ways to do a correct-looking analysis of the data to get evidence to support your agenda. Maybe your agenda is right and maybe it's not, but I'd love to hear a story of someone standing up and saying "your consultant submitted a report with glaring flaws, they should not be paid and you should reconsider X." It's more likely the little company just goes out of business or the big company buries the failure.
  - ifyoubuildit 4y ago
    So whats the alternative? "Just trust me"?
    - yunwal 4y ago
      The alternative is to trust in most circumstances (you did hire me after all), and do thorough analysis once in a while as a gut check.

_jx7j 4y ago

>Meaning you could absolutely suck at your job or be incredible at it and you’d get nearly the same regards in either case. One of the things I don't like about statements like this said in a Data Science context, is that they are true outside of Data Science as well. Executives make big decisions, managers make smaller decisions, nobody can evaluate how good/bad they really were for months or years. Engineers build something amazing, or build a house of cards, nobody cares as long as the money people are happy, even if the business use case turns out to be wrong in the long run. >With a short-term focus they also won't really care, because they can still put these results in marketing materials and impress most outsiders as well. Forget Data Science, you see this in KPIs as well. Say a crappy metric has to be moved by Q2 next year and people will destroy the company to move it. I feel like Data Science is just one of those areas where you are exposed to a wider range of people and get to feel the full crapola of the insanity of working in a corporation. For lots of roles (e.g. Engineering) you get to hide in a hole behind layers of people and not see some of this insanity.

scottLobster 4y ago
Not to get too off topic, but as a 35 year old engineer it seems the world in general has far fewer consequences than I was raised to expect. Everything from businesses with bullshit ideas flourishing at a loss, to January 6 even being possible (politics aside I expected the Capitol Police to crack a lot more skulls than they did once people started smashing windows), to the whole FTX situation and the tepid response in the media/government, to petty crime being outright tolerated, to in my own career I've at times burned through enough money badly enough (albeit with good intentions) that I thought I was going to be fired, only to be told in a performance review I was doing a good job (grateful to stay employed but WTF, I would have fired or at least demoted me). Importantly, the motivation for this lack of consequence doesn't seem to stem from a desire for forgiveness or positive reinforcement or any mechanism that might make things better. It seems like there's a general apathy/nihilism that's growing in society, whereas by contrast my entire education from childhood up I was held to strict standards and reliably punished when I failed to meet them, and this was in US public schools (albeit a highly ranked school district) and a public university. That or I was just raised in a bubble, and the historical examples I referenced growing up and reference to this day are just a case of survivorship bias, and all the bullshit that was alongside them back in the day has simply been forgotten. I'm not sure, but it is disappointing how little people at large seem to give a shit. Maybe it's a side-effect of the obesity epidemic and people just have less energy or something

2devnull 4y ago

This gave me a chuckle. If you read the feature article you understand that this is also because management wants “decision driven data.” They have an idea and use ds to provide charts and tables to support their idea. The harder the idea is to support, the greater value data science is able to provide. I guess data science is inferior to research in this way. People care about research methods, rigor, etc… Maybe data scientists should adopt stricter standards, like actual scientists.

samvher 4y ago
I did read the article - some of the problems with judgements of work quality also come up with (hypothetical) well-intentioned truth-seeking non-political long-term-optimizing managers who just don't happen to be stats experts.
- 2devnull 4y ago
  Sorry, wasn’t trying to imply you didn’t and I fully agree. Even managers that know stats can be busy or but into hype about ml or other shiny new things that they don’t have time or resources to deconstruct. This is another big problem with data science, “black box” systems and cargo cults. It’s easy to think “LLMs will change the world! We should use them, the competition will.”

dqpb 4y ago

> Approaches that are wrong from a statistics point of view are more likely to generate impressive seeming results. This is to be expected from an information theory point of view. It's why "fake news" will always be a thing.

jmount 4y ago

Super point. I can't resist repeating it back. When incorrect work outperforms correct work in superficial evaluation, it is then selected for.

mmcnl 4y ago

This is somewhat captured in the article as well. "Managers will say they want to make data-driven decisions, but they really want decision-driven data. If you strayed from this role– e.g. by warning people not to pursue stupid ideas– your reward was their disdain, then they’d do it anyway, then it wouldn’t work (what a shocker). The only way to win is to become a stooge." In science, a good scientific result can be bad for business. There is often little appreciation for the "science" in data science.

throwingit0 4y ago
>There is often little appreciation for the "science" in data science. It feels like even Google falls prey to this at times: they keep redoing the same A/B test until it comes up in favor of the change (or the designer whose pet project it is runs out of political capital, presumably).

whiplash451 4y ago

You should pay the price for data leakage very quickly in production. Does management look at slides or AB test dashboards?

disgruntledphd2 4y ago
People don't pay attention to production metrics, and a noisy problem (like marketing or whatnot) can often be pretty bad for a looooonnnnnggg time before anyone notices.

jrochkind1 4y ago

> Approaches that are wrong from a statistics point of view When OP talked about "the main bottleneck to my work" in terms of areas he would need to learn more about -- I was expecting him to talk about facility with statistical methods and using them appropriately! I'm not sure what to take from the fact that he never did! I would like to ask him what he thinks about that!

mumblemumble 4y ago

The problem is that nobody actually wants data science. They want data pseudoscience. And for the same reason that people tend to want pseudoscience instead of science in any other domain, too. Science is slow, tentative, and messy, and usually responds to questions with even more questions rather than with answers. Pseudoscience tends to be much more concerned with exuding confidence and providing clean-cut answers. It's what happens when a desire for science meets a need for instant gratification. Along the way, things like blinding and controls and watching for bias and validating assumptions tend to get dropped when they're inconvenient or difficult to explain. And they're always inconvenient and difficult to explain.

onlyrealcuzzo 4y ago
> The problem is that nobody actually wants data science. They want data pseudoscience. Technically, I think investors & owners would want the company to use real data science to improve products & maximize profits. Everybody in the middle just wants to use data to lie to get promoted faster - because you don't get promoted for actually doing a good job - you get promoted for convincing people you did a good job, and lying is a VERY useful / effective tool.
- deckard1 4y ago
  Welcome to the world of OKRs/KPIs/Scrum, where everything's made up and the points don't matter.
- NeutralCrane 4y ago
  > I think investors & owners would want the company to use real data science to improve products & maximize profits. This is based on the assumption that companies are focused on long term profits and stability, and I’m not sure why anyone believes that to be the case anymore. The vast majority of companies are run based on next quarter’s stock price or growth metrics. I worked on a newly formed data science team coming out of grad school that was tasked with taking some predictive initiatives that the company had relied on external consultants to produce, and implementing them in-house. The external team’s results always looked exactly like what the business wanted to hear, but they rarely played out in practice. This was in part because the underlying data quality was terrible, and the company wasn’t executing in a way that allowed anyone to actually answer the questions being asked. The consultants would just torture the data until they could come up with a report that would ensure the company would come back the following year. So we spent a lot of time trying pouring cold water into the business groups who saw data science as a magic wand that would conjure up more money at no cost. But we never were able to convince them to invest in anything that would take longer than a year. Anything that would require a change in their marketing or strategy executions that wouldn’t immediately deliver increased results was just a non-starter. But actual data science requires that kind of investment for long-term layoffs. So the data science team became figure-heads, never given the buy-in to actually make impact on business, but kept around so teams and leaders could tout being “data-driven” and throw “AI” and “machine-learning” into PR and marketing materials. You aren’t wrong about middle management is looking to get promoted faster. But every single individual from the employee looking for a promotion to the executive suite to the investors are addicted to incentive windows no longer than 6-12 months.

rgrieselhuber 4y ago

I’ve always disliked how data science was positioned within companies as well, it’s outside the critical path of product and engineering, which means it becomes a mere abstraction to management (e.g. “throw that problem to the data science team and see what they come up with”), resulting in very vague and abstract requirements and, hence, deliverables. I think there is huge value in the discipline and technologies, but it unfairly gets relegated when not integrated to the whole product / engineering process. Hence, the title / concept of Data Engineer seems like a much better fit for this role within many companies.

rgavuliak 4y ago
Yeah, as a Data Science manager I've experienced this pain a lot (not part of the critical path). I am now an Engineering Manager that works with a cross-functional team including FE/BE/DS/Devops and it's the most power I ever had to put Data Science in front of our clients in a meaningful way.

hcks 4y ago

> Because these issues are subtle, management will often not pick up on them or not be aware that this kind of thing can go wrong. Management is not your teacher at school, it is not there to check up your results make sense. Management mostly assumes you’re competent at your job.