perfect-is-the-enemy-of-good

Apparently attributed first to Voltaire. Dans ses écrits, un sage Italien / Dit que le mieux est l'ennemi du bien.

Though he seems to attribute it, himself, to another.

I quite like that quote. What brought it to your mind, @skullpatrol?

@Aurora0001 mentioned it above: chat.stackexchange.com/transcript/message/42294388#42294388

It is also one of my guiding principles, actually.

Also see "Worse is Better" by Dick Gabriel: en.wikipedia.org/wiki/Worse_is_better

It is used as a partial justification for some of the agile software principles.

@BenI. the latest (scratch v python) question seems tailor made for an answer from @Heather. Maybe you could ping her to see if she is interested.

One worry about "level" based tags is that "beginning programming", for example, can be taught at several levels. I think that that (as a tag) is more relevant than "HS" or "undergraduate" or any other general educational level indicator. I think we had that tag and gave it up. I think that was a mistake.

@BenI. for example, teaches many things that not every undergraduate learns, though he teaches HS.

@heather As per Buffy above, "the latest (scratch v python) question seems tailor made for an answer from Heather." You might be interested in checking it out ;-)

The problem with introductory-programming was that it applied to >50% of the content on the site. And, if you apply it as "introducing some programming concept", >90%. And to further the problem, introductory programming in middle school is vastly different than introductory programming for adults, like no comprende would engage in. The tag just didn't add any value.

Interesting new question. Something different for us, I think.

Yes! And it deserves a meta, because it's an xx-ideas sort of question, so now that one of the -ideas tags has been eliminated, we need to agree that we are applying that concept (no list closure) based on human judgement; if it's looking for classroom ideas, it won't be given a list closure.

I think we will be lucky to get any answers at all, actually. We probably have users who can answer, but I don't think the regulars are among them.

And proven wrong, just like that. Great answer @Aurora0001

Well, I must admit, machine learning isn't actually something I'm hugely familiar with... but I dabble with it occasionally. Never made anything particularly effective with it though

I tried to make a classifier a while ago, and to my delight, it was more than 90% accurate. Until I realised that I'd trained it with too much of one class and not enough of the other. After balancing the two classes, the true accuracy was barely more than 50%, so essentially random guessing was nearly as good.

But I did learn a valuable lesson: that treating it as a black box does give you working code, but means that you don't have a clue how to improve the accuracy other than "more data"

That said... "more data" is essentially the approach of a lot of the successes in ML

There is an article in the Dec 18-25 Issue of the New Yorker: The Numbers King, about big data and the Flatiron Institute. More about the people than the thing itself, though. I think it is probably online, but don't know the availability. A library can probably give access to any academic.