search

thomasp85 / lime

Local Interpretable Model-Agnostic Explanations (R port of original Python package)
https://lime.data-imaginist.com/
Other
486 stars 110 forks source link

Question about LIME results #182

Open ms1948 opened 3 years ago

ms1948 commented 3 years ago

Hi, I am trying to use LIME on R, and I need some explanations about the results produced by the package. I have 144 records with 25 parameters each. Records are labeled “Cat” or “C3” depending on the parameter's value. If I run a LIME analysis with the command: model <- train(b_train, b_lab, method = 'rf') two times with the same test record I receive two different results (see annex).
Please can you explain to me this behaviour ?

S001A S001B

lingminhao commented 2 years ago

I guess that's because of the instability of the LIME framework. https://towardsdatascience.com/instability-of-lime-explanations-3e0efc00a7de Personally, I think if your predictions have no noise then they should be stable. Not sure about this though.