Convert three different csv into a graph and do classification?

[monk1337]

Hi, I have three different csv files, for example data

I want to use the qualified column as a a label and predict based on the variables given in data_dm and data_cm csv files. How to convert these three CSV into graph along with labels and do classification?

Thank you!

[rusty1s]

It doesn't look like your data naturally converts to a graph since tables are only connected based on a 1:1 relationship. Instead, you can flatten these tables together based on payer_id and perform logistic regression/XGBoost/MLPs to make predictions per row.

[monk1337]

Thank you quick reply. This is demo data, and the actual data contains many columns and hierarchal medical records. Could you share any guidance on how I should proceed to convert such data into a knowledge graph and make node predictions?
My goal is to convert data into a knowledge graph and do node/link classification. But I couldn't find any example of how to convert such data along with labels?

Thank you!

[rusty1s]

You may find this tutorial helpful. Let me know.

[monk1337]

Thank you for the tutorial; I have already gone through the tutorial, and I have two questions:

1. How to load labels in that tutorial and do node classification?
2. How to use the same tutorial if data has many columns and relationships?

[rusty1s]

1. Instead of adding all features to the features, you can select the label column and add it as an separate attribute to your data, e.g., data[node_type].y = torch.from_numpy(df["my_label_column"].values).
2. The tutorial should be agnostic to the number of tables/columns, i.e., you call load_edge_csv for every PK/FK relationship.