The one you’re most comfortable with is the one we recommend. The main thing is to be able to keep it up to date.
1 Like
We are keeping the setup guides here updated:
5 Likes
Is it the notion of broadcasting alone that is responsible for eliminating the need for loops when doing matrix multiplication?
I think a[i] should be not ar dimension but ac dimension (I think Jeremy misspoke).
The loop is still there, it’s just written in a faster language. At the end if the day, the fastest version, a CUDA kernel, will still have those loops, they will just be done extremely fast in c and in parallel.
6 Likes
Also thanks Rachel for the “rows and columns” song, that’s probably going to prevent more bugs than any test suite 
4 Likes
Don’t trust Jeremy. Only trust Rachel’s song 
6 Likes
I didn’t make it up, just learned it from others. There are more lyrics, but I can’t find them right now
6 Likes
Those rules are really really helpful !
1 Like
Got the following error when I opened the notebook nb_00.py:
FileNotFoundError: [Errno 2] No such file or directory: '/usr/local/opt/python3/bin/python3.5': '/usr/local/opt/python3/bin/python3.5'
How do I update fastai library?
Let us know when you do, I need another song for my 4 month old to fall asleep to.
I think I need to go through the code in detail but my understanding from what Jeremy just went through is that the dimensionality trick (unsqueeze) + broadcasting is what allows linear algebra libraries (in any language) to do matrix multiplication without for loops. Of course I might have misunderstood haha
Where can I catch up on the parts of the lecture I missed?
You don’t write the for loop, but those libraries aren’t magically computing the result. They do the for loop in your place.
sudo conda install -c fastai fastai
I might be wrong but it seems you don’t have python installed or your python executable is not found in the PATH. The error says it couldn’t find the python executable.
Verify your python installation.
To install/update fastai check the repo: https://github.com/fastai/fastai
Rewind, then watch on whatever speed you desire (I’m a fan of 1.5x)
Is it broadcasting alone that reduces the time complexity of the matrix multiply operation?
The complexity is the same, the time is down because broadcasting allows you to use a faster language (C in this case) behind the scenes.
2 Likes
I don’t think it improves the time complexity, just the computation time.
1 Like