Thanks for the report. A fix is welcome

Desperately waiting for first time issues tag XD

You're welcome to take it on as a first time contributor. My main reason for not adding that tag is that this may not be trivial to write a test for.

@jnothman, if no one is looking into this, would you mind clarifying the kind of tests necessary here?

Do we want to assert that calls to base_estimator.fit include the correct sample_weight (when provided to RANSACRegressor.fit) via mock or pytest’s monkeypatch?

Or would it be sufficient to give RANSACRegressor a dummy base_estimator that requires the sample_weight argument in its fit method and ensure it runs without catching on fire?

It is not enough to pass sample_weight to base_estimator.fit(), because it is the loss_function within RANSACRegressor.fit() who determined whether the weighted inlier_subset is good enough. Sample_weight should be passed to loss_function as well.

@jnothman, this class has been made private, is this issue still open though?

As in should it really still be open?

This is what I have so far. Can someone please check this? @jnothman, @gerhajdu?

J-A16@e0d78ce#diff-15e1ad3883740ee06b2d4e15f86cad68

It passes pytest test_ransac.py, by the way. No errors. Inside that test file, there are weights passed to the fit method of RANSACRegressor. I did a couple tests where I edited RANSACRegressor to make it fail, but the code I posted works.

@jnothman, this class has been made private, is this issue still open though?

The class is not private. It is publicly available for import from linear_model. Only the module is private.

Oh, first time dealing with private anything in Python, actually. Thanks for letting me know.

I thought since it was in a private file, maybe it was.

Do you think I need to change my code at all?

If you have code, please submit a pull request for review.

I agree with @pallashadow that it looks like the loss calculation should also be weighted... but we ideally drive this from tests. See #15657