Anchor explanation with regression

[Navdeep Gill]

• Have you looked into ways to explain regression models using Anchors? Would binning help in this case?

[Marco Tulio Correia Ribeiro]

Yes, and yes. I've used anchor to explain functions on the output of regression models, such as:

1. If X and Y, Output > 100
2. If Z , 100 < Output < 120

I think using one sided binning functions (like 1 above) is more interpretable than two sided ones (like 2 above). In example 2 above, Z must be a very specific condition, as it can't lead to outputs smaller than 100 OR larger than 120. This makes for anchors that are too specific.

[Navdeep Gill]

Thanks @marcotcr! Will look at this further.

[Manu Joseph V]

@marcotcr @navdeep-G Although I understand how binning would help with this, I am not entirely clear on how to implement it in Anchor.

Is there any example or a notebook which handles Regression?