Random fashion thoughts - Part II (A New Hope)

[LA Guy]

This isn't a meaningful distinction. If I am a prominent flat earther activist I would be both standing against the societal consensus viewpoint and representing the consensus viewpoint of my constituency. I don't believe Cathy represents the consensus of data scientists and institutional consumers of these models in terms of her activism towards transparent models. If it were, Google wouldn't be making photo recognition algorithms that labeled black people as gorillas, and we'd know how credit scores worked. Most people just don't talk or think about how these algorithms impact peoples' lives, which makes her work valuable.

I should have been more clear. Her axioms are those of "liberals", which is about half the US.

Like I've said, she is technically impeccable. However, her politics highly color her assessment of her industry. She is quite open about this. It's in the book. But I think that it detracts from what she writes. Ymmv.

 

[the shah]

This isn't a meaningful distinction. If I am a prominent flat earther activist I would be both standing against the societal consensus viewpoint and representing the consensus viewpoint of my constituency. I don't believe Cathy represents the consensus of data scientists and institutional consumers of these models in terms of her activism towards transparent models. If it were, Google wouldn't be making photo recognition algorithms that labeled black people as gorillas, and we'd know how credit scores worked. Most people just don't talk or think about how these algorithms impact peoples' lives, which makes her work valuable.

are you suggesting that data scientists are opposed to transparency?

the gorilla gaff is due to inadequate training data (not taught what gorillas are and so it started generalizing based on darkness, etc), and credit scores and interest rate offerings, while not entirely simple, aren't exotic ideas they are just probabilistic models that don't extend as well as they should to a large and diverse population.

 

[King Calder]

FICO is literally a black box algorithm that companies who use it are contractually forbidden from trying to unlock or reverse engineer. Most of the data scientists I know from school and work with professionally are douchebags who either think their models capture objective truth or are openly disdainful of their clients who are "too dumb to understand their work." Many I know *hate Cathy O'Neil.

To be more precise, I think there's a lot of data scientists who are pretty arrogant in the pursuit of their work, and for every one who works at google and pushes their code on git there's another two working in like intelligence or education consulting who's building a black box for their PE owners to sell to municipalities or school boards with zero oversight or accountability.

 

[King Calder]

Data science is a trade like any other. You get hired to build **** and most don't ask a lot of questions, just like how engineers get asked to build everything from bridges to bombs.

 

[conceptual 4est]

Every time the SF is dying thing comes up, people always mention Instagram being the new cool place. Who should I be following? Because at the moment with:

@penanceroyaltea I just get pictures of delicious home cooked meals and wine. It's appalling
@conceptual 4est I just get pictures of vinyls on a Friday evening and homemade ice-cream
@Synthese I just get pictures of dogs and some more dogs
@Fokken I just get pictures of Belgian loft apartments and cats
@KamoteJoe I just get pictures of concerts I'm not cool enough to go to
@diniro I just get pictures of bbq and indian takeaway
@the shah I just get moody pictures of train stations

You should follow No Man Walks Alone!

Next up is fresh pandan ice cream. I found some leaves at a market here

 

[LA Guy]

FICO is literally a black box algorithm that companies who use it are contractually forbidden from trying to unlock or reverse engineer. Most of the data scientists I know from school and work with professionally are douchebags who either think their models capture objective truth or are openly disdainful of their clients who are "too dumb to understand their work." Many I know *hate Cathy O'Neil.

To be more precise, I think there's a lot of data scientists who are pretty arrogant in the pursuit of their work, and for every one who works at google and pushes their code on git there's another two working in like intelligence or education consulting who's building a black box for their PE owners to sell to municipalities or school boards with zero oversight or accountability.

To be fair, Cathy is in open warfare against the industry, and she is opening disdainful of most of them, whom she characterizes, and I think often accurately, as "data analysts", who don't really understand the underpinnings of the models with which they work, and are mostly plug and playing, and as "data engineers", who, like you said, build models, but are fundamentally either uninterested in, or unable to, study data science itself, which is what a data "scientist" does. If you think of yourself as a data scientist, and a smart (smarter than you) colleague downgrades you to a tech, you would be disinclined to like them.

As for Git - I find that it's both informative and incredibly frustrating. For example, I've been using it to try to figure out what has worked to stop the rogue ads that are propagating through the net. The answer is "We've tried a ton of things, but very little actually works." Which, if I were still an academic, would just be fine. But instead, I have people from here and different websites complaining about them (and rightfully so), but with no understanding of what is going on and the level of sophistication of these bad players. I have the fun choice of either spending my time solving the problem, or trying to educate someone who really just wants a solution and will not be satisfied if one is not immediately available. So I'm on Git yelling (mostly at myself), saying "Please, so many bright people, no answers at all. OMG!"

Not even going to discuss what we have implemented, since it seems to be working for now, and if I let the cat out of the bag, well, who knows what people will think of next.

 

[John L]

Has this been posted?

If not, I feel like this should be posted.

 

[the shah]

To be fair, Cathy is in open warfare against the industry, and she is opening disdainful of most of them, whom she characterizes, and I think often accurately, as "data analysts", who don't really understand the underpinnings of the models with which they work, and are mostly plug and playing, and as "data engineers", who, like you said, build models, but are fundamentally either uninterested in, or unable to, study data science itself, which is what a data "scientist" does. If you think of yourself as a data scientist, and a smart (smarter than you) colleague downgrades you to a tech, you would be disinclined to like them.

this is a real problem as there is no clear definition of a data scientist and what distinguishes one from a data analyst ("oh you must be really good at excel!"). there is at least some differentiation with data engineers (who build data pipelines, something that was born out of the need for specialisation in moving data around so the data scientists can do their gorilla maths on them).

the term science is a bit of a misnomer as well, and the fact that it's a hot field leads to tons of folks pitching themselves as data scientists and in many organisations hiring managers, never mind HR, can't see through the BS. and then you get charismatic tableau bros (tabros is an amazing term) vying for some of the pie as well. and really it's market demand being met by the opportunistic. which leads to the criticism offered by the would-be usurper king calder -- there's only one shah in this town.

i haven't read her book and will get to it eventually, but I think a more fundamental issue is tackled in this text:
https://www.amazon.co.uk/Innumeracy-Mathematical-Illiteracy-Its-Consequences/dp/0606265953

 

[Baron]

I should have been more clear. Her axioms are those of "liberals", which is about half the US.

Like I've said, she is technically impeccable. However, her politics highly color her assessment of her industry. She is quite open about this. It's in the book. But I think that it detracts from what she writes. Ymmv.

Yeah, from what I remember from that econtalk episode, I thought her orthodoxy undermined her position. Especially in conversation with Russ Roberts, who is open about his own classical-liberal biases, but he's also expansively fair-minded and humble about his opinions and makes a real effort to find common ground with ideological opposites. If you don't engage him on that level, you end up looking very petty.

 

[Benjaminba]

Current SWD challenge ending Saturday, so there's still a little time to put together an entry!

https://www.styleforum.net/threads/...e-oct-18th-nov-1st.586057/page-3#post-9072576

 

[cyc wid it]

Hey if you're a Stylefarmer who is a machine learning researcher or practitioner, come work with me

 

[OccultaVexillum]

Recruit me Cyc!

 

[LA Guy]

Hey if you're a Stylefarmer who is a machine learning researcher or practitioner, come work with me

Aren't machine learners the new programmers? Don't you guys get recruited by Amazon and Alibaba for NBA money?

Hell, I just need a good dataminer and analysts/scientists to tell me a few things about this database.

 

[oulipien]

FICO is literally a black box algorithm that companies who use it are contractually forbidden from trying to unlock or reverse engineer.

OTOH if someone could reverse-engineer FICO the bureaus would probably be interested in knowing about that, because even they don't really understand, necessarily, why a particular history results in a particular score—lots of these models are black boxes to the people who create them, too. Not for nothing is ML called the high-interest credit card of technical debt.

 

[sinnedk]

Soo PSA folks, i needed to send some money via paypal in foreign currency and i opted for my credit card which has no foreign conversion fee to do the conversion instead of paypal, the conversion came out almost 1:1 with the rate i get from google whereas PP was $25 more...