bknyaz / graph_attention_pool Goto Github PK

Hello!

Is the threshold-based pooling method in your paper pretty much the same as using the "min_score" argument in pytorch geometric's TopKPooling?

Thank you for sharing this repo with us!

Could you please add an explicit LICENSE file to the repo so that it's clear
under what terms the content is provided, and under what terms user
contributions are licensed?

GitHub docs on licensing

However, without a license, the default copyright laws apply, meaning that you retain all rights to your source code and no one may reproduce, distribute, or create derivative works from your work. If you're creating an open source project, we strongly encourage you to include an open source license. The Open Source Guide provides additional guidance on choosing the correct license for your project.

Thanks!

Hi, I think this is a valuable work. I want to know how to create my own dataset as the format of mnist_75sp_train.
For example, I write a Pytorch DataLoader as follows:
data_loader=DataLoader(my_dataset,batch_size=batch_size,shuffle=True)
It returns batch_images (shape: batch_size*H*W) and batch_labels (shape: batch_size*1) in each batch. How should I do to construct my own dataset and graph for the train? Can you provide a code to help to do this? Thank you very much!

I am trying to generate the mnist75 dataset by running: ./scripts/prepare_data.sh and I am getting the following stacktrace:
Fr Dez 16 17:57:00 CET 2022
start time: 2022-12-16 17:57:01.936994
dataset mnist
data_dir ./data
out_dir ./data
split train
threads 0
n_sp 75
compactness 0.25
seed 111
/home/mada/anaconda3/lib/python3.9/site-packages/skimage/_shared/utils.py:338: FutureWarning: multichannel is a deprecated argument name for slic. It will be removed in version 1.0. Please use channel_axis instead.
warnings.warn(self.warning_msg.format(
Traceback (most recent call last):
File "graph_attention_pool/extract_superpixels.py", line 128, in
sp_data.append(process_image((images[i], i, n_images, args, True, True)))
File "graph_attention_pool/extract_superpixels.py", line 55, in process_image
assert n_sp_extracted == np.max(superpixels) + 1, ('superpixel indices', np.unique(superpixels)) # make sure superpixel indices are numbers from 0 to n-1
AssertionError: ('superpixel indices', array([ 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34,
35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51,
52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68,
69, 70, 71]))

Do you know what the issue might be?

Thank you in advance!

Hi @bknyaz, thanks for the well-documented codebase and new graph datasets!

I wanted to confirm: are the GNN models for the MNIST dataset operating on the fully-connected graph or a k-Nearest Neighbor graph (as in the MoNet and ChebNet papers, which use k = 8)?

To me, it seems the code is using dense graphs for now.