Question about dataset

Questions
#1

Hello, I want to know How do I use my own data set

1 Like
#2

Can you elaborate a bit more about your scenario?

  • What task is this dataset for?
  • What’s the format of your dataset?
  • Do you have one graph or multiple graphs?
  • Do you have node/edge features?
1 Like
#3

I want to know if i have a new dataset,no preprocessing, and i want to put it into our model, Is there a tutorial for pre-processing data to help us process the data into a form that can be entered into the model

#4

Currently we don’t have something like that. If you have a particular dataset that you want to work on, you can share more information on it.

#5

Ok thank you very much

#6

Hi mufeili,

Imagine I have many small graphs (approximately 1000 nodes for each graph) and I have all nodes labelled or annotated in all graphs. In this case, I assigned 1 (out of 6 labelled) to each node. Along with the label, I do have 5 features (2 binary features with Yes or No, and 3 multi class features with range between 0-20).

For the edge, I have undirected edges in each graph with 2 features weight (float) and feel (binary integer).

My aim is to do predict an unlabelled/unannotated graphs after training. Does node classification is a what I need to look at? or do you have any suggestion? a simple implementation would help me a lot.

Thanks all

#7

It sounds like you want to perform graph classification. See if this tutorial helps.

#8

Thanks for the reply and I’m sorry I typed my question wrongly. What I meant is to do node classification. I come across GCN, but it seems its implementation takes one big graph instead of multiple disconnected graphs as I have.

So, could you perhaps recommend a suitable technique for me?

Many thanks

#9

You can use dgl.batch to batch multiple disconnected graphs into multiple big graphs. Essentially you are combining the batched graph training in graph classification and the node classification example.