1,124 karma · joined May 3, 2017
In my experience, the most difficult parts of the process are (1) translating from qualitative problems “in the field” to a formalized technical problem, and (2) all the wrangling necessary to implement the formal problem using field-collected data.
I find that I spend most of my time as a data scientist working on these parts of the process, and often don’t have the bandwidth or even requirement for more advanced methods. Not that fancy techniques are the goal, per se, but I do notice I rarely have the opportunity to use them.
That's really interesting! Can you explain it a little more? I can't read Chinese, so it's a bit opaque what's going on...
The "problems" you describe are just...nonexistent, and really not a big deal. I have literally never encountered confusion about our marital status, problems with forms, or difficulty being introduced.
Why not just use an offline-first format like docx (collaborative: comments and track changes), synced with Dropbox or similar (sync across devices)?
Also, my use case is 100% non critical from a security standpoint, so I can afford to be careless... your point definitely stands in sensitive environments
It is decidedly not true that I want to update my venv for every minor version bump!
I deploy to a cloud service w/ a specific version; my package manager is slow to update; I develop collaboratively using a shared container w/ a fixed version.
Updating shared resources with every new release is not always realistic, and that makes it so that I do need (or at least want) to use a virtualenv.
Get this guy a dictionary!
I meant that it can serve the role of low-level, nitty gritty machinery, entirely separate from its use case as a module unto its own right.
Eg, pandas (in some sense) is just a convenience layer on top of numpy—but to me that’s like saying any piece of software is “just a convenience layer on top of python.” It’s partly a means to an end! Not just an end unto itself.
Numpy enables a whole new class of functionality, independent of its direct use as part of my software. Is there a better word for that?
Being in love does feel earth-shatteringly special, unimaginably unique, irreplaceable; but the author is right that that’s how everyone in love feels.
I don’t think that cheapens love, or means that my love for my spouse is “less” because others feel symmetrically the same, or even because I could have felt it with someone else, too.
I think that is what it is to be in love: an overwhelming all-consuming feeling of everything being one-of-a-kind and precious and Right. Thank God it’s not impossibly rare, or only possible under the most specific circumstances!
(This is not to say that love is easy or feels like that all the time—it takes a lot of work, but that work feels worth it and important because of how it feels during the moments when it does come effortlessly.)
It is interesting to me that in many respects it serves as “guts”. I definitely drop down to use numpy directly with regularity. But it’s also possible to do 80-90% of the job without ever explicitly using the module itself. It’s baked in everywhere, to the point that it feels like just another standard library.
Exciting to see progress! Keep up the good work, numpy team :)
To some degree, this may be "just" aa matter of a poor training set: too few / unbalanced examples. Can't we just use a better dataset?
But this also reflects a more systemic problem with descriptive / learned behavior, rather than ontological representations: we will always have these dataset problems, to be fixed after the fact when someone points out the flaw. See also facial recognition of non-white faces.
We want intelligent systems that can learn from the world, but that's not an accurate mental model of how humans learn. Often, we learn our values and develop opinions in spite of empirical representations—I don't read the newspaper to find out what Muslim people are like, I start from my values / convictions (all men are created equal, every person deserves respect and the benefit of the doubt prima facie) and maybe update on the basis of observations of media depictions.
Yes, a computer may be able to imitate human speech after reading a lot of text...but that doesn't mean it understands the content of that text.
Does that mean that if you have both a pacemaker and an iPhone 12, you would still only know about this if you sought out the support page?
That's very reactive, and not so helpful—I would be more likely to go to a hospital if my medical device failed, than to a consumer electronics support page...
> a software engineer or a scientist who wants to become a machine learning engineer or a data scientist
and earlier in the page it says
> you will be ready to build complex AI systems, pass an interview or start your own business
I think that's misleading, and just not true.
:(
Eek. This type of extreme claim makes me even more skeptical than I already was of such a short overview. Each of the topics listed could easily fill 100 pages themselves...
All who needs to know? Hopefully not anyone actually using this in their job. Maybe it would be a useful reference for the definition of the term, but it's completely wild to suggest you'll have any understanding of SVM, NN, CV, or gradient descent after only 100 pages, much less all of them...
But also a really fascinating technology! I wasn’t aware of photosensitizing, what a novel way of inside and outside used in conjunction
1. ML is used to improve UX / ops 2. Other people's jobs depend on the quality of the ML output
1 can encompass things like offering recommendations or optimizing click-through. 2 is things like analytics-as-a-product in smart manufacturing.
I assume 1 is a lot easier to "succeed" at: if I receive a bad music recommendation, I just move onto the next one.
2 is harder to do well, and the output quickly loses credibility when it's bad.
That said, bad ML in high-stakes contexts (by which I mean, someone else is using the ML to do their job better) should get weeded out pretty quickly...so for the total effort and care taken to address 2, they may be more successful on average.
I’m disappointed to see this only applied to a “toy” example...the real value is proven (or not, as is often the case) when the “rubber meets the road” and is applied to real problems on the factory floor. A POC like this is a nice demo, a fun exercise, but a long way from “productionized and used by real manufacturers.” The lack of a real example suggest to me this hasn’t actually been used in the field...
I’m also curious how one would communicate a very technical/abstract model like this to existing practitioners of eg Six Sigma. My experience has been there is a lot of friction in introducing new methods in such an old-school setting: not only do old habits die hard, but quality engineers are often hired explicitly to do particular six sigma analyses. The difficulty is in convincing a client that these methods (a) can be useful and reliable in the context of traditional methods, and (b) aren’t intended to replace their role. I’ve seen the most traction w methods that connect directly to established SPC procedures, while also pushing the limits of these known and _interpretable_ techniques.