Chain of encoding: efficiency issue

kdd-lab / LORE_sa

Code for LORE (stable and actionable)

3 stars 1 forks source link

Chain of encoding: efficiency issue #9

Closed rinziv closed 4 months ago

rinziv commented 10 months ago

https://github.com/kdd-lab/LORE_sa/blob/b24451e8efbbf1dcac819a5389ea11e8dadf50e6/lore_sa/encoder_decoder/tabular_encoder.py#L15

It seems that encoding a new record requires creating a new set of instances of the sub-encoders: every time a new OneHot is created, a new LabelEncc, etc. Maybe they can be stored and reused, since the dataset descriptor they use is not changing

rinziv commented 4 months ago

this section has been completely rewritten