Cannot run TabDiff if the dataset does not contain `int_col`

[ghabault]

As the title suggests, if someone attempts to use TabDiff with a dataset that does not contain an int_col (resulting in the process_dataset function printing "Int 0" at the end), the training will fail with the following errors:

• AttributeError: Can't pickle local object 'preprocess.<locals>.<lambda>'
• pickle load EOFError: Ran out of input

This issue may need to be addressed in the implementation to enhance the usability of TabDiff. Or at least mentioned in the readme under the subsection Using your own dataset from the Data Preparation section).

[YuanfengZhang]

It's not a big deal, I guess. You can write yourself. I have made one in my fork for generalization. Maybe you can use it.

[ghabault]

The implications might be more important than I initially anticipated. For instance, I used a dataset that originally did not contain an int_col. To avoid previous issues, I created a column with only the integer 0, assuming TabDiff could handle such a redundant column. While I can train and test with this revised dataset (using --mode train and --mode test flags), I cannot use classifier-free guidance (--y_only flag).