Model comparison metrics

[ashwanijha1]

Hi,

I'm trying to evaluate forrest diffusion model performance as you did in your great paper, but on my own datasets (to determine the value to altering hyperparameters etc).

Do you have a vignette/ example where you calculate:

• reconstruction metrics such as Wasserstein-2
• imputation accuracy (following manufactured missingness)
  On a train/ test/ generated datasets consisting of mixed continuous, categorical and binary data please?

I know that it is in your 'script_generation' code somewhere but I am finding it challenging to adapt it / make it work with my own dataset.

I'm working in python.

Thanks so much,
Ash