packtpublishing/graph-neural-networks

Hands-On Graph Neural Networks Using Python, published by Packt

graph-neural-networks's Issues

PeMS-M dataset download

Thanks for this book, but I have problem with PeMS-M dataset in chapter 15. Can you update the download link?

installing book environment

I can't get to install the exact version of pytorch.
ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'pytorch'
note that I found python 3.8.10 not 15

Anyone managed to have the environment working with newer or same version of the libraries?

I think I found a typo in the book. (p127)

Thank you for writing such a great book and helping me understand GNN better.

While reading a book, I think I found a typo in the book.
At 127p, I understand Figure 8.2 to be an explanation of embedding through neighbors in Figure 8.1.
In Figure 8.1, the 1-hop of node (1) is represented by [0, 4, 6], while in Figure 8.2, the 1-hop of node (1) is represented by [0, 4, 5, 6].

The Amazon link is wrong.

The current Amazon link redirects to a different book.

What does node2vec expect in .wv method?

Hello,

I;m working through chapter 4. In the prediction of the test set, you use train_mask_str in the book but in your gh code you use train_mask. Depending on which one you use one gets very different results.

if you pass the integer mask you get something that won't go pass 60% accuracy.

y_pred = clf.predict(node2vec.wv[test_mask])
acc = accuracy_score(y_pred, labels[test_mask])
print(f'Node2Vec accuracy = {acc*100:.2f}%')
>>>Node2Vec accuracy = 59.09%

But if i pass the string mask i get what you report both in gh and the book . Which seems really high considering how many samples you pick out in the train and test.

y_pred = clf.predict(node2vec.wv[test_mask_str])
acc = accuracy_score(y_pred, labels[test_mask])
print(f'Node2Vec accuracy = {acc*100:.2f}%')
>>>Node2Vec accuracy = 95.45%

How can one explain this discrepancy?

Chapter 01 and 03 Karate Club Data accuracy(training and test) for 4 classes

Working through the excellent chapters to expand my python expertise. Chapters 01 and 03 have sections that process the
Karate Club data using 2 different methods. I would like to compare the training and prediction accuracies for both procedures.
The chapter 01 GNN analysis is based on 2 classes (Mr. Hi vs Officer) while the wordvec2 procedure in chapter 03 has 4 classes.
While the test accuracies for the 2 class case are ~98%, however when I apply the 4 class data, the training accuracy drops to 59%.
I am not sure whether my approach is correct.

In short, could an analysis of both procedures be provided for the Karate Club data with 4 classes? This would address differences
between GNN and wordvec2.

Thanks,
DGC

Typo in Figure 7.2

Thanks for this awesome book!

Here I find a typo:

In Figure 7.2 (P108), the last $a^{1}_{ij}$

should be $\alpha^{i}_{ij}$.

Also for $a^{2}_{ij}$.