Awesome! Very nice channel in which 2 apparently different topics such as graphs and XAI are outstandingly explained! It would be very nice to combine both of them to lead to a better graph algorithmic fairness
@@DeepFindr yeah! I've already seen it and its really good! It was very useful to deepen in graph XAI libraries and research group that I didn't know. Im staying tuned because all the videos absolutely freak me out!
Hi, please could you tell me if it is possible to perform node classification on a predicted set of values for each node? Just like what you showed us for Cosa dataset (for static graph), but now for this temporal GNNs?
Amazing video! I have a question regarding one data point attributes. We have both x and y. The features themselves are the speeds which is the ground truth right? We want to train our model to predict into next 12 timestamps and compare it with values in x. So what is the significance of y (though in documentation, its given that y is the ground truth, both x and y should have the same values but they differ. Only difference that has to be is that x additionally has time of day as a second feature for each of the 12 timestamps per data point) Could you kindly clarify this?
Hey, I wonder why the model gives different output every time? I tried to use sMAPE as an evaluation metric, and the result was different every time I ran it. Do you have any idea why that is?
@@florianro.9185 I think it says that its StaticGraphTemporalFeature data type or something ,becasue i tried just using len() on it. Although I need to double check later
What could be the reason that the produced error i.e MAE is high in this example. The actual MAE is very low for the same dataset and same topology. Furthermore, how you can bring the results into datasets original rang(0-70).
Hi! You can find all options in the documentation: pytorch-geometric-temporal.readthedocs.io/en/latest/modules/root.html If you need to do more modifications I recommend to copy the source code and adjust it: pytorch-geometric-temporal.readthedocs.io/en/latest/_modules/torch_geometric_temporal/nn/recurrent/attentiontemporalgcn.html#A3TGCN
Hello, Thanks for the great video. I tried following this for one of my projects using GATNet(Pytorch). When trying to give METRLA data, I face an error saying, "Static Graphs not allowed." I tried changing the class, but the error still persists. While chickenpox does seem to work, I was confused about how both differ. I am trying to use my own data, which is much similar to METRLA. Is there a way to contact you to send screenshots? or any comments you have that can help? Thanks
@@DeepFindr Hey, Thanks for the reply. Yes, I did see the video but as I explained the aspect of giving n-features and n-targets throws that error, I am trying to understand.
Thank you.. This video is very helpful. Just a quick query - I am getting this error "ImportError: cannot import name 'container_abcs' from 'torch._six' (/usr/local/lib/python3.7/dist-packages/torch/_six.py)" at this step "from torch_geometric_temporal.dataset import METRLADatasetLoader" , tough this was working fine before. How can I resolve this ?
Hi! I had a similar problem recently. Have you tried running this cell at the beginning of the notebook: !pip install torch==1.8.1+cu111 torchvision==0.9.1+cu111 torchaudio==0.8.1 -f download.pytorch.org/whl/lts/1.8/torch_lts.html !pip install torchtext==0.9.1 Downgrading to torch 1.8 solved it for me for now. (github.com/qubvel/segmentation_models.pytorch/issues/424)
Thanks for the helpful material. I wonder about the graph-level regression ( estimation) application via GNN. Could please direct me to your materials(if any) or other sources such as papers, or tutorials that include coding examples for that specific task?
For the a3tgcn block this option is unfortunately not available (at least I didn't find it). This is something I mentioned in the video. For other layers such as GConvGRU you can simply stack multiple ones and also choose the hidden dimension. I hope there will be more support for customization in the future. :)
![[LIVE] : ONE ลุมพินี 89 | คู่เอก "ยอดไอคิว vs คิริลล์"](http://i.ytimg.com/vi/Z3vYkHJ_5Ig/mqdefault.jpg)
Awesome! Very nice channel in which 2 apparently different topics such as graphs and XAI are outstandingly explained! It would be very nice to combine both of them to lead to a better graph algorithmic fairness
Thanks a lot for your feedback!
I have a video called "how to explain GNNs" - not sure if you have seen it :) it's about XAI + GNNs
@@DeepFindr yeah! I've already seen it and its really good! It was very useful to deepen in graph XAI libraries and research group that I didn't know. Im staying tuned because all the videos absolutely freak me out!
Haha I'm very happy that you like it :)
Hi, please could you tell me if it is possible to perform node classification on a predicted set of values for each node? Just like what you showed us for Cosa dataset (for static graph), but now for this temporal GNNs?
Thanks a lor for the great video and explanation, awesome 😍
Hi, where I can find a tutorial to learn how to create my own dataset to use as input of a TGCN? Thank you!
Did you find one?
Amazing video! I have a question regarding one data point attributes. We have both x and y. The features themselves are the speeds which is the ground truth right? We want to train our model to predict into next 12 timestamps and compare it with values in x. So what is the significance of y (though in documentation, its given that y is the ground truth, both x and y should have the same values but they differ. Only difference that has to be is that x additionally has time of day as a second feature for each of the 12 timestamps per data point)
Could you kindly clarify this?
Hey, I wonder why the model gives different output every time? I tried to use sMAPE as an evaluation metric, and the result was different every time I ran it. Do you have any idea why that is?
Thanks for this video series. Is there a way to get a prediction of underlying interaction graph( the adjacency matrix or edge index)
I encountered this problem recently. Why the code that len(set(dataset)) doesn't work? It always show unhashable type: 'Data' error……
That sounds to me as if your Data object is not a Dataset. Have you put your data into a dataset class?
@@DeepFindr even i am getting this error.
@@DeepFindr yeah set(dataset) doesnt work for me either
What happens without set? Can you share the output of type(dataset)?
@@florianro.9185 I think it says that its StaticGraphTemporalFeature data type or something ,becasue i tried just using len() on it. Although I need to double check later
I have trouble finetuning the model?? has anyone managed to recreate the results from the paper?
Thanks for the material. i ams wondering what do you do with the 0 values in dataset, just leave as it is or interpolate them?
What could be the reason that the produced error i.e MAE is high in this example. The actual MAE is very low for the same dataset and same topology. Furthermore, how you can bring the results into datasets original rang(0-70).
How can I tune the model? What are some of the hyperparameters that can be tuned for A3TGCN?
Hi! You can find all options in the documentation: pytorch-geometric-temporal.readthedocs.io/en/latest/modules/root.html
If you need to do more modifications I recommend to copy the source code and adjust it: pytorch-geometric-temporal.readthedocs.io/en/latest/_modules/torch_geometric_temporal/nn/recurrent/attentiontemporalgcn.html#A3TGCN
Hello, Thanks for the great video. I tried following this for one of my projects using GATNet(Pytorch). When trying to give METRLA data, I face an error saying, "Static Graphs not allowed." I tried changing the class, but the error still persists. While chickenpox does seem to work, I was confused about how both differ. I am trying to use my own data, which is much similar to METRLA. Is there a way to contact you to send screenshots? or any comments you have that can help?
Thanks
Hi! Did you see my tutorial on building temporal graph datasets?
@@DeepFindr Hey, Thanks for the reply. Yes, I did see the video but as I explained the aspect of giving n-features and n-targets throws that error, I am trying to understand.
Can I predict next 12 timesteps but more than one target feature with this model?
Amazing!! Thank you so much =]
Thank you.. This video is very helpful.
Just a quick query - I am getting this error "ImportError: cannot import name 'container_abcs' from 'torch._six' (/usr/local/lib/python3.7/dist-packages/torch/_six.py)" at this step "from torch_geometric_temporal.dataset import METRLADatasetLoader" , tough this was working fine before. How can I resolve this ?
Hi! I had a similar problem recently. Have you tried running this cell at the beginning of the notebook:
!pip install torch==1.8.1+cu111 torchvision==0.9.1+cu111 torchaudio==0.8.1 -f download.pytorch.org/whl/lts/1.8/torch_lts.html !pip install torchtext==0.9.1
Downgrading to torch 1.8 solved it for me for now. (github.com/qubvel/segmentation_models.pytorch/issues/424)
@@DeepFindr Thank you , this resolved my issue.
Great :)
Thankyou so much !
Great video!
Thanks for the helpful material. I wonder about the graph-level regression ( estimation) application via GNN.
Could please direct me to your materials(if any) or other sources such as papers, or tutorials that include coding examples for that specific task?
Hi! In my GNN Project series, the third video (or fourth) is about graph-level predictions :) hope that's what you are looking for.
Thanks, interesting video.. How would you tune such a model?
Thanks!
You mean how to change hyperparameters? Or what do you mean? :)
@@DeepFindr Yes if more layers need to be added, or what parameters have most impact on loss? :)
For the a3tgcn block this option is unfortunately not available (at least I didn't find it). This is something I mentioned in the video.
For other layers such as GConvGRU you can simply stack multiple ones and also choose the hidden dimension.
I hope there will be more support for customization in the future. :)
installation is very difficult
👏🏻👏🏻❤
can you talking about HGCN?
Hi! I'll put it on my list but cannot promise as I first have to do the fraud detection video and some others :D :) thanks for your request
@@DeepFindr thanks. i have question about pytorch geometric 2.0. if i installed PyG 2.0, whether i also installed graphgym?
Hi! What exactly do you mean? :)
op
TLDR: "couldn't make it work but maybe you can"
Well done.
Sir,
Kindly share your contact details for those who have enquiries privately.
Thank you