18:35 In the step where the edge indices are normalized, don't you think there will be a weird behavior because it might happen that there are isolated nodes in the graph. Then the edge_index and node_index would not match and in the message passing step, the message from one node will propagate to an incorrect node because the adjacency information is now changed
Hi, hope you're doing well.... You said some information about Frankenstain dataset, you said images are replaced instead of atoms in molecule... I've used Ensyms dataset from network repository, but there isn't any information about it.... Of course there are some statistical information but there isn't any about, for example what does the label of each node show? or something like this.... Would you please help me? I really nead it's information. Thanks in advance
how can I convert normal dataset used normally with sklearn to a graph .. I couldn't catch this.. for instance: I have a dataset with features like the table below f1,f2,f3,f4, result 1,3,44,5,6, 1 2,7,56,4,12, 0 2,7,56,4,12, 2
Does your dataset have an underlying graph structure? If yes, you can create an edge list for rows in the table that are connected to other rows. Then you will have a sparse binary adjacency matrix which you can input in pytorch or pytorch geometric. The features here will correspond to the node features and the "result" column has the node labels.
As far as I understood, the Data Id's are from range 0-(n-1), due to pandas series indexing and the way that css is encoded, but they should be indexed from 1-n the index+=1 is just a cheap trick using pandas implementation
![[LIVE] : ONE ลุมพินี 89 | คู่เอก "ยอดไอคิว vs คิริลล์"](http://i.ytimg.com/vi/Z3vYkHJ_5Ig/mqdefault.jpg)
Wonderful, it's help me a lot, thank u
18:35 In the step where the edge indices are normalized, don't you think there will be a weird behavior because it might happen that there are isolated nodes in the graph. Then the edge_index and node_index would not match and in the message passing step, the message from one node will propagate to an incorrect node because the adjacency information is now changed
Thank you
Hi, hope you're doing well....
You said some information about Frankenstain dataset, you said images are replaced instead of atoms in molecule...
I've used Ensyms dataset from network repository, but there isn't any information about it....
Of course there are some statistical information but there isn't any about, for example what does the label of each node show? or something like this....
Would you please help me? I really nead it's information.
Thanks in advance
My graphs was created from images and saved as pt file how can i apply graph mining later please with this format
how can I convert normal dataset used normally with sklearn to a graph .. I couldn't catch this..
for instance: I have a dataset with features like the table below
f1,f2,f3,f4, result
1,3,44,5,6, 1
2,7,56,4,12, 0
2,7,56,4,12, 2
Does your dataset have an underlying graph structure? If yes, you can create an edge list for rows in the table that are connected to other rows. Then you will have a sparse binary adjacency matrix which you can input in pytorch or pytorch geometric. The features here will correspond to the node features and the "result" column has the node labels.
13:15 what is the purpose node_attrs.index += 1 similarly others edge_index.index, graph_idx.index, etc.
As far as I understood, the Data Id's are from range 0-(n-1), due to pandas series indexing and the way that css is encoded, but they should be indexed from 1-n the index+=1 is just a cheap trick using pandas implementation