Incompatibility between SynthCity and scikit_learn 1.4. With workaround and possible fix

[MarkJacksonRG]

Description

I think there's an incompatibility between synthcity and the new 1.4 version of scikit_learn?

How to Reproduce

I tried running the 'load and run a tabular generator' bit from the docs and got an error.
OneHotEncoder.__init__() got an unexpected keyword 'sparse'

Click to view screenshot of code and error:

Expected Behavior

It should fit the model without an error

Workaround

The OneHotEncoder docs say sparse was renamed to sparse_output. The new parameter was introduced in scikit_learn 1.2 and the old one removed in scikit_learn 1.4.

I fixed my local installation by downgrading scikit_learn to 1.3.2.
pip install --upgrade scikit_learn==1.3.2

Possible fix

I checked the stack trace and I think it might be possible to fix it with a one-line change. I haven't actually tried it! I'm not setup as a contributor.

Change sparse to sparse_output in line 63 in tabular_encoder.py : I.e. this line

cat_encoder_params: dict = dict(handle_unknown="ignore", sparse=False)

I thought I would flag it here in case it's something the community want to fix. I think you'd need to require scikit_learn 1.2 and greater.

System Information

• OS: Windows 10
• Python 3.10

[robsdavis]

Hi MarkJacksonRG, Many thanks for your issue! This has now been resolved in the manner you described. This can been seen in #253. If you should ever want to contribute in the future it is very easy (see CONTRIBUTING.MD)!