Working with huge Hetero Data

[mprithiv]

Hi

I currently have a huge hetero data am trying to build a HeteroData object using this link. The number of nodes and features are huge that the object creation takes for ever. I am using SentenceTransformer for ahpha-numeric features, LabelEncoder for numeric and date features. Is there any way to build the graph more efficiently.

[rusty1s]

Can you say with part of the pipeline takes the most time? I think this mostly might be due to the usage of SentenceTransformer (whose efficiency is out of our scope).