#3 PyTorch Book Thread: Sunday, 12th Sept 8AM PT

Note: This is a wiki, please edit it & add resources!

:tv: YouTube Link: PyTorch Book Reading - 3. Train your first neural network - YouTube

Hi all!
This thread is for discussing, Q&A, and everything else for the #3 meetup of the book reading group.

We’re ready to train our first Neural Network this week! We’ll look at how to use torch.nn, understand NNs and start looking into convolutions

:round_pushpin:Link to sign up


<<<< Previous Session Thread


Q: When are we training our first torch model?


today only … :grinning: isn’t it ?


Explain * operator in Python. I think it would be helpful for the beginners.


Yes, * as a mathematical operator does elementwise multiplication, ** does exponentiation and @ does matrix multiplication.

Additionally * can also unpack a list to pass arguments to a function


It was about function argument which means putting * in function argument will allow accessing a number of arguments on runtime.


Thank you for the PyTorch series covering the Deep Learning with PyTorch Book.
I wanted to know how to coordinate between the fastbook (Jeremy Howard’s material) and this series?
Also should we pick fundamental papers like those by LeCun and Hinton or stick with interpretation of their concepts in recent works?
While implementating papers using the submissions at paperswithcode, many repositories are in TensorFlow, how to deal with that?

1 Like

Are there any reputation score or some such to be able to edit the wiki?

I cannot seem to be able to find any “Edit” button.

1 Like

I wanted Sanyam to explain that in the video. He did.

1 Like

in between triple dot and link there is a pencil, use that

I try to learn from Jeremy’s fastbook by keeping the book and code only colab notebook side by side, try to modify and learn about the code, and then watch the lecture (as warching the lectures take time without knowing the code and approach.
How do you recommend tackling with PyTorch and Fast.ai (part 1)

train_dataset_pytorch = torchvision.datasets.FashionMNIST(root='./data/', train=True, transform=transforms, download=True)

This is not working as I learned that this dataset has been removed. Any solution about how to load this dataset in pytorch? Or any solution about how to load dataset from an url in pytorch? I searched for it but couldn’t understand their solution.

Sometimes there are four sets as well. The fourth one’s often called dev-test.

1 Like

I can’t access docstring on jupyter notebook. What shortcut keys are you using?

Radek also mentions reading papers in his book Ultralearning. I recommend reading it.

1 Like

I asked about the papers only with the intent of implementing them, however older papers are in TF or caffe in that case what should be done?

Not to mention that a lot of papers are glorified brute force architecture search.

In one session coukd you cover the updated approach of using EC2 as an alternative to google coLab (pr example when data augmentation is to be done for a huge data set but locally)

1 Like

Jeremy had very briefly covered this

There are much more papers on Papers with Code written in PyTorch than the ones in TensorFlow.

You can always look for indie implementations. There are always one or two lying around.

Use Google.