pfnet-research / TabCSDI

A code for the NeurIPS 2022 Table Representation Learning Workshop paper: "Diffusion models for missing value imputation in tabular data"
MIT License
42 stars 4 forks source link

min-max normalization before training #6

Open kuiyuanshao opened 2 months ago

kuiyuanshao commented 2 months ago

Hey guys,

I am not quite sure why min-max normalization is being used across your scripts and also mentioned in you guys' preprint. Compared to the original paper CSDI, they used z-score normalization which guarantees the result after normalization has mean zero and variance of one, and the min-max normalization cannot guarantee this.

I am wondering what would be the purpose of changing it, did you guys find better performance with min-max?

Regards, Keiran.