Pre-processing code

[ericmjl]

Hey @ericmjl, yep we’re getting there 😁 ! Docs are on the horizon.

The plan for tomorrow is:

1. Docs coverage (as close 100% as sanity allows)
2. Feature Pre-processing

Re: pre-processing I was thinking of a setup where users can pass a few standard functions (one-hot, mean, etc) to a dictionary/config as well as fitted sklearn scalers for normalising across graphs in a dataset (with some helpful functions for creating them from a list of graphs) or unfitted sklearn scalers (for single graph normalisation). A config object seems nice and consistent, but a dictionary might be better for users that create their own features as a config object would be a bit inflexible there.

Would be super keen to hear any thoughts/suggestions on this.

pre_processing_dict = {
“molecular_weight”: StandardScaler,
“secondary_structure”: partial(one_hot, vocab = SS_ELEMENTS)
}

G = process_graph(G, pre_processing_dict)

After this, I think the only outstanding task is a conversion toolkit to support various frameworks. Then it’s cleaning up & polishing before I think a V2.0 release is in order :D
EDIT: and tests! Will crack on with them this week.

Originally posted by @a-r-j in #45 (comment)