working with custom csv data

[chillum-codeX]

working with custom csv data
i have 12 features and have to predict seriousness

[rusty1s]

What's the question? :)

[chillum-codeX]

Whenever a patient takes a vaccine and experiences any adverse event then he reports it and a case ID(VAERS_ID) is created.

Nodes like symptoms experienced, other medicines he took, serioudness of the event(LTHREAT : life threatening situation) etc are considered with repective relationship as shown in the image.

Our aim is to predict whether a case will lead to life threatening(LTHREAT) situation or not on the basis of symptoms experienced, vaccine taken and other medicines the patient took.

Here nodes like symptom, other medicines, vaccine and Lthreat doesn't have any features and edges are unweighed. So how will we define the nodes without features and unweighed edges.

[rusty1s]

Thank you. While I think one can interpret this as a graph, isn't it more naturally to encode these features (symptons, age, gender, ...) into a single feature vector and utilize traditional machine learning with IID samples?

Alternatively, you can interpret each symptom, etc as a node in your graph, and learn an embedding for that node in an end-to-end fahsion jointly with the prediction task, e.g., via torch.nn.Embedding. Happy to chat more about this.

[chillum-codeX]

Instead of using nn.embeddings to define a node without features is there any other way to define it

[rusty1s]

Yes, there exists. One alternative way is to define node features by some graph statistics, e.g., node degree. We provide support for this via transforms.OneHotDegree or transforms.LocalDegreeProfile.

[chillum-codeX]

Thanks @rusty1s
i will try and let you know