Hi Community!
I have a spare-time project where I am currently encountering some mind issues. I am not a professional AI/ML master, but I did try to do some online courses about it at least. However, I stumbled upon the key term "Graph Neural Network" which fits (in my opinion) my problem very very good. Nevertheless, since this topic is completely new to me and I dont know where to start, I am hoping that somebody could give me a helping hand here.
I try to break the problem itself down to the basics.
I have X Nodes and each has a feature vector with length N. Each node can take a different representation/approach which I simplified to integers(0-5). So Node 1 can be of type "3", node 2 can be of type "5", node 3 of type "2" etc.
The graph itself leads to multiple result values. My approach here is to summarize the result values and give the result a point range for evaluating its quality. So 0 points means bad solution, 10 points means good solution.
What I want to do now:
After training the NN using my training data, I want to build a graph and predict the output value for each combination of node representations(e.g. Node 1 uses approach 2, Node 2 uses approach 0 etc...)
Exemplary Figure:
![[Image: systems2.png]](https://i.ibb.co/TYRFGLL/systems2.png)
What I dont know:
Everything :(
Best practice to structure input data
The way how to implement the node representation integer (maybe as another column in the feature vector? or is it possible to directly label the node?)
Setup of GNN (GNNs havent been part of the courses unfortunately)
Thank you in advance for any hints, advices and help
I have a spare-time project where I am currently encountering some mind issues. I am not a professional AI/ML master, but I did try to do some online courses about it at least. However, I stumbled upon the key term "Graph Neural Network" which fits (in my opinion) my problem very very good. Nevertheless, since this topic is completely new to me and I dont know where to start, I am hoping that somebody could give me a helping hand here.
I try to break the problem itself down to the basics.
I have X Nodes and each has a feature vector with length N. Each node can take a different representation/approach which I simplified to integers(0-5). So Node 1 can be of type "3", node 2 can be of type "5", node 3 of type "2" etc.
The graph itself leads to multiple result values. My approach here is to summarize the result values and give the result a point range for evaluating its quality. So 0 points means bad solution, 10 points means good solution.
What I want to do now:
After training the NN using my training data, I want to build a graph and predict the output value for each combination of node representations(e.g. Node 1 uses approach 2, Node 2 uses approach 0 etc...)
Exemplary Figure:
What I dont know:
Everything :(
Best practice to structure input data
The way how to implement the node representation integer (maybe as another column in the feature vector? or is it possible to directly label the node?)
Setup of GNN (GNNs havent been part of the courses unfortunately)
Thank you in advance for any hints, advices and help
