schneiderkamplab / syntheval Goto Github PK

in get_cat_variables in variable_detection.py in the utils folder, it checks if the column dtype is object, then it appends it to the cat_variabes list, which is the list of categorical features of my dataset. I want to ask why does it also check for dtypes int64 and float64? Aren't those supposed to belong to the list of numerical features which is the num_cols list?

Hello! Firstly, thank you for creating this wonderful library!

I had a question regarding the DCR privacy metric implementation.

In your paper and code, you compare the median DCR of (synthetic, train) and (train, train) distances.

I was wondering if you considered comparing the median DCR of (synthetic, train) and (holdout, train)? I have usually seen this version (page 4 in ref), and I was wondering if you had any insights into why (train, train) was used.

I am using the DCR metric in a project and wanted to know if a holdout dataset should be used or not.

May I ask if there is any guide or documentation on how to interpret the metrics? For example I am trying to figure out weather a higher score in Nearest Neighbour Adversarial Accuracy represents higher similarity between my synthetic and real data, and vice versa.
Or is it possible to provide a list of references to the original proposals of the metrics?
Thank you!