Does Jeremy’s formula for Frobenius norm assume only square matrices (since there is the same n for both i and j? Shouldn’t it be n and m for rows and columns numbers?
3 Likes
It does, it was just an example.
1 Like
I love this. Jeremy promised to teach us advanced deep learning and is instead turning us all into an army of super low level numerical programming wizards. Genius idea to improve the whole ecosystem - there are nowhere near enough people who can do this stuff.
The ol’ bait-and-switch… but I am not complaining!
To be clear - I’m sure we are going to learn a heck of a lot of advanced machine learning during the course - more “advanced” than we ever thought - just not the way we were thinking.
10 Likes
Speaking of LaTex, I cheat by using https://www.codecogs.com/latex/eqneditor.php which generates an image for me like so:
26 Likes
the course reminds me this article on how to use GPU do GEMM efficiently.
https://devblogs.nvidia.com/cutlass-linear-algebra-cuda/
3 Likes
The m are the individual elements of the matrix.
2 Likes
Can you broadcast with text arrays? Or only with tensors?
Is there any difference for broadcasting in numpy and pytorch?
Mathpix is also a pretty handy tool which can go the other way (it converts an image into LaTeX).
13 Likes
Is this why we see in some notebooks (like Rossmann)
y_range = torch.tensor([0, 1.2], device=defaults.device)
instead of
y_range = [0,1.2]
Because we are then tapping into something faster than python?
1 Like
PyTorch does its best to imitate numpy, so not that I know of.
1 Like
Partly, yes. Also, you can’t broadcast a list with a tensor that has multiple dimensions. It has to be a tensor too.
1 Like
That’s so helpful. I hate writing unsqueeze. The “None” notation will make my code more readable.
6 Likes
And I’d never quite understood what that [None,:] step was doing apart from “changing the rank somehow”.
1 Like
I’m jealous of Jeremy’s Excel-foo skills. Are there any recommend sources for high-end excel use that anybody knows of?
5 Likes
Will this excel file be posted too (although it’s a simple one)? The excel files helped me a lot in part 1.
I think Excel knowledge is best acquired “just in time”. Better to try using it, think “hmm I wonder how you do X in excel”, google it, and try. Excel is so flexible that the chances of finding a guide that directly helps you is pretty slim.
2 Likes
Platform question — I hope this is the right place!
Is there updated advice regarding what platform is the most straightforward to use run jupyter on a GPU-enabled machine and keep up with the course notebooks?
In Part 1, many platforms were referenced in the beginning but some of them (e.g., straight AWS) seemed to fall out of compatibility as the course proceeded.
So which platforms are advised for Part 2?
3 Likes
You suck at excel by Joel Spolsky is a pretty good primer for programmers.
13 Likes
Haha hopefully I stand corrected 