Friendly Introduction to Temporal Graph Neural Networks (and some Traffic Forecasting)

[@me_hanics]

Thank you! This channel seems like it's run by someone well-read. I very much love that you include research papers too.

[@henk_iii]

Thank you for the excellent presentation of the topic, a job well done!

[@eneserdogan34]

Amazing video on such a challenging topic

[@leo.y.comprendo]

Great first video on TGNNs! Looking forward to the next part

[@cuongnguyenuc1776]

This work is crazy good!!

[@mukeshnarendran1083]

Great video and am enjoying the GNN playlist. If there are open requests for new topics will it be okay to request for a hands on tutorial for skeletal action recognition (ex: berkley MHAD dataset) using temporal-based graph networks. Thanks

[@prakaashsukhwal1984]

awesome explanation. Do share some more pytorch based example use-cases on temporal GNNs. Appreciate your sharing!

[@DeepFindr]

Hi! I have one for traffic forecasting :) and a video on how to generate temporal graph datasets

[@zahramovahedinia1896]

This video is wonderful! Please, I wish to know what is the difference between higher and lower proximity orders? (In the table shown at 11:55)

[@chientruong926]

Thank you so much!!!

[@shrutishrestha8296]

Great content. Thank you.

[@nobywils]

Amazing explanation sir!

[@avishkarsaha8506]

great tutorial on TGNNs

[@mountainwolf12]

Dear instructor, is it possible to do node classification on temporal GNNs as well? for example, using a trained model predicting the category of each node during each time step based on given time series of node features?

[@sunnyarora4916]

My CSV dataset has time values and corresponding sensor values measured every 5 min, when i tried to make a graph out of this TS ,taking node the time value and sensor value its attribute. Node gets wrong sensor values. How shall I convert this CSV to correct Graph? Any basic codes, I am a beginner?

[@minalpatil564]

very nicely explained..

[@wilfredomartel7781]

Nice work!

[@wijdanchoukri6138]

Great content, very helpful Thank you so much

[@DeepFindr]

Happy that you found it useful :)

[@mountainwolf12]

Hi, can node classification be done on these final temporal GNN outputs, for a predicted timestep in the future for each node?

[@vincentyang8393]

Thanks, Very helpful!

[@kylecherry9869]

Thanks, really nice intro.

[@DeepFindr]

Thx!

[@HeduAI]

Awesome animations! Which software did you use to create them?

[@DeepFindr]

Thank you

[@ayushsaha5539]

I don't understand the gating mechanism and the math behind the time series predictions. Any recommendations (i.e. papers)?

[@VinayKumar-sd4dh]

how do we handle the case where graph changes over a period of time? I mean the graph at t1 will not be same as graph at t2.

[@DeepFindr]

This is called dynamic graph and for example explained in this paper: arxiv.org/abs/2006.10637

[@DeepFindr]

The important part is to only update nodes that are present in the current graph. But nodes might be added or removed over time

[@VinayKumar-sd4dh]

@@DeepFindr thanks for the paper. Your videos are really helpful and simple to understand.

[@armandogomis7732]

Thank you very much for your video!!! I look forward to the next part. I am currently working with elliptic dataset to create temporal based classification of a dynamic graph with pytorch geometric, but I am finding it impossible to adapt the dataset to pytorch geometric. I would be very grateful if you could show me how to create a custom temporal dynamic graph in the library because the toy examples don't help much. Regards.

[@DeepFindr]

Hi! I somehow didn't receive a notification for your comment but found it by accident :D Next part will be uploaded in the next days. I will say some words around datasets and hope this will help you with your issues. Otherwise feel free to leave a comment and I'll come back to you :) I agree the examples could be better :D

[@priyankagautam4932]

can we use dgl for the GNN -lstm framework ?

[@DEBRAJDEiot]

Thank you for the very helpful video! Is the mentioned Google Colab notebook available to test and play with.

[@DeepFindr]

Very soon :) comes with the next video in the next days! Sorry for the delay

[@DEBRAJDEiot]

@@DeepFindr Thank you! :)

[@DEBRAJDEiot]

No problem at all! :) Thank you for sharing information and knowledge to the technical community