Did I mention I think Allen is totally great yet?
Did I mention I think Allen is totally great yet?
1. The data he fits is for distance (10k), and the conclusion ("why you can't beat the Freeze") concerns sprinting.
2. The claim that speed is a multiplicative model is fine and I agree based on the 10k data. But the people running against the Freeze aren't uniform from the population, so the argument is missing a step about what distribution (what percentile of the background distribution) they might be from.
3. There are some interesting numbers about how the high school girl is x% faster than Downey, the boy is y% faster than the girl, etc, but they don't link to the stats later in the article.
Of course I might be wrong (maybe people in my general area are just incredibly slow)? But I’d have some confidence in the assertion anyway.
My eldest is 10. He is good at math because we do math games all the time. I can’t get algebra to sink in properly yet. Doesn’t really click in the time we have and with the teacher he has (me). But in Python he is giving variables a name all the time. Perhaps there is something about programmatic math that helps, not hinders education.
On the hinders side though, I can understand it being confusing to try to grasp:
y = x + 2
As being sort of 'constantly updating', referring to x in abstract, when you've just learnt that Python will assign (once) the result of computing x (at that time) + 2 to y.You could def y as being x() + 2, but you might just be delaying the confusion to when you try to introduce functions in math and find yoursel saying 'no no that was just a hack to get python to behave more like algebra'.
Is there any language that allows this I wonder? Where you can do something like:
y = &(*x + 2)
That is, y is a pointer to the result of adding 2 to the value pointed to by x, and this is somehow kept up to date, by tracking references like garbage collection perhaps, or at compile time by inserting the code to update y whenever x is updated. (Names would need to be globally unique I suppose, or else x transformed in to some kind of struct that could contain y and other dependents I suppose, and then also a link from y to x so you can remove that if y definition changes.. but that's a lot of load for the humble assignment operator!)It might be tempting to think it's procedural vs declarative, but it's not really, you could procedurally evaluate that y should now be maintained to have such value. Admittedly you can't really do it any other way on declarative terms. But what allows you to do computation like that? Datalog maybe?
I'm ignoring Matlab/Octave/etc. because I'm not that familiar and of course (I assume) they work more like math. What's interesting is that everything else seems to do it differently; I'm not sure why or how that happened. Is it as simple as an early limitation (in terms of early computation and compilers) that stuck?
You could, I guess, in any language that allows operator overloading (Python would be a good one) define operators that actually return a function for computation, guess. This just seems like a really complicated way of defining a function, though.
Or possible a template library like Eigen could be thought of in terms of “building a computation.” Maybe CUDA has something like this? I’m not sure, at all.
An aspect for my case specifically, which is probably not a coincidence, is that I usually work with Electrical Engineering students. They’ll typically have seen some very light Verilog in some of their intro classes. I’m definitely not a Verilog expert, but you can use it to define combinatorial circuits—in that case you are literally describing the wires and gates that the computation flows through, so the left hand side does change based on the right hand side, continuously. At least until you write to a clocked register. And of course in a real circuit, there’d be some signal propagation time…
Seriously though, Allen is an instructor. These are educational pieces to bring a spark of understanding and creativity to students. I don’t think he is aspiring on modeling the absolute truth in these posts.
This piece has a learning about the working of non-normal distributions in sampling and reasoning about cut-off points.
No model is perfect, but I don’t have the slightest clue why you think that would follow.
Also, Ethiopian runners would be a counterargument. It's just reading way too much in a model that probably was there before the data. Seeing how everybody defends it as "learning", the example was sought to be described by the model.
Wat. His model says that many factors have to be perfect for someone to be as fast as Usain Bolt. If anything, it implies the opposite of what you said