Captum TCAV train.py sgd_train_linear_model() Where do the weights come from?

[FilipUniBe]

I posted this originally on StackOverflow, but now I thought it may fit better here. After all it is related to this setup.

I am using the captum setup https://captum.ai/api/concept.html https://captum.ai/tutorials/TCAV_Image

In the Default Classifier the linear classification boils down to the function sgd_train_linear_model(). When I debug the code until sklearn_model.fit(x, y, sample_weight=w, **fit_kwargs) at line 327 the variabe "w" is empty.

But then at line 338:

Convert weights to pytorch

classes = (
    torch.IntTensor(sklearn_model.classes_)
    if hasattr(sklearn_model, "classes_")
    else None
)

It can suddendly derrive the weights from the model. How so? Help me understand.

Debug and read the docs.