What is the meaning of the values of the "blue" features in the lime output? I understand that they influence the lime black-box model to classify an observation as 0 label, but what is the meaning of their numbers?

For example the next figure:

Lime results. Retrieved from towardsdatascience][1]

What does represent the value 0.20 of the variable alcohol? Is the coefficient for this feature in the simple local model (regression for example) that the lime is using is -0.20?

There are serval results about this topic, but I haven't found an answer to this question.

  • The figure retrieved from towardsdatascience article written by Radečić
  • $\begingroup$ I asked myself the same thing and I'm not sure, but I guess you get it right. LIME is a feature attribution method and from my understanding summing up the intercept and each of the 'explanations' multiplied by the actual feature value should result in prediction of the surrogate or explanation model. I have some examples in which it works like that and others in which it doesn't. The latter might be caused by a surrogate model that doesn't perform well... $\endgroup$
    – So S
    Apr 22, 2022 at 19:09


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.