Maybe Jeremy will mention it later but: as far as I understand, Jeremy is planning to do new videos to cover things he didn’t have time to do in the 8-14 lessons (like audio, image detection?).
If it’s still current, how will that work?
It is based on the course, so if you want to provide comments, we’re all ears
Hmm, certainly happy to give feedback, but if you have a work in progress, the feedback might be more useful. Otherwise I’ll have to guess what you’ve put in the book from the lectures, notebooks, example code, forum threads, etc.
Hope object detection will also be covered
Yes, we are still planning to have additional lectures (which will be livestreamed for you), but the exact details, including dates, have not been set yet. We will keep you posted!
… and seq2seq (attention, transformer, transformer-xl, etc…)
@AlexisGallagher got a great talk on value semantics
Is right to think a reference as “an alias” ?
Yes, that’s exactly right. References are aliases (or multiple names) to the same “thing”.
The parameter names in, out, and inout come from/are used in modula2
Feels like Swift makes it pretty hard to do mutation unless you’re very intentional about it (and this seems like a very good thing)
Why will y
change if it uses let
for creation? Shouldn’t it be immutable object?
So if you wanted to append 92 on all of them, would you have to loop through all of the arrays?
OK there is an answer.
Wouldn’t you just define it as a struct not a class?
I think you can use a pointer to them instead
What’s great about Swift is that in particular it makes mutation of shared mutable state hard, since it is reference semantics that allow two different variables to point to the same location and thus share the value at that location.
If you declare a variable with a value semantic type as a let
, you can’t mutate it. If you declare it as a var
, you can mutate it, but you still are safe from anyone else mutating it via another variable, because access to it is not shared with another variable.
Could you please remind how this S4TF autodiff is related to the “usual” Swift? It uses a custom version of the compiler, right? So we don’t have this thing in the language itself (yet)?
It’s in the tensorflow branch of swift, so not in the official swift but in S4TF.
Thinking about this a bit more, I don’t think the “IDEs mean this is an old fashioned problem” answer is compatible with using Jupyter as an IDE, at least until Jupyter has a “jump to def” function In the meantime I’ll trust Chris that it’s mostly a comfort thing and “you’ll get used to it”.