[BUG] MultipleASE and OmnibusEmbed mistakenly have a .transform method

[bdpedigo]

Expected Behavior

We never tested or intended the out of sample transform to work for those multiple graph embeddings as the inputs wouldn't even make sense

Additional Details

• Either of these classes could have a transform method if someone wants to implement it and go through the math
• For now, I recommend that we just override this method in the BaseEmbedMulti class and have them raise a not implemented error

[rajpratyush]

@bdpedigo can i take up this issue??

[bdpedigo]

@rajpratyush sure

[loftusa]

@bdpedigo

Are we sure we want to override transform? That prevents us from calling fit then transform separately on our graphs. Usually you wouldn't want to in practice, but I don't see why we should prevent it (since sklearn API allows for it)?

Might be better to check specifically if the new transform inputs are out-of-sample, and then raise NotImplementedError for that. But that is also trickier to implement.

[bdpedigo]

@hugwuoke possible thing to work on