Showing results for 
Search instead for 
Did you mean: 

What is the strength of an explanation?

Linear Actuator

@Eu Jin @IraWatt 


As I understand it ...



A Data Robot prediction for a binary classifier is a certainty of classification score between 0 and 1. The certainty that the row denotes a positive case. A corresponding explanation is a triple (f,v,s) giving a feature, its value, and the strength of the impact.


I have heard it said that the strength is something like the rate of change of the certainty with the value. But, given two features f1 and f2, which have a big step impact and a small affine impact, f1 might have the smaller strength in that sense, but be the more important feature in determining the conclusion. 


So, in this case - in explicit Data Robot terms - which has the highest strength? and why?


What does the strength given tell you about what is the best thing to try changing to increase the probability.



0 Kudos
1 Reply
Community Manager
Community Manager

Closing this thread because it's duplicated over here, where everyone is chiming in on this very interesting topic. Thanks for bringing this to the Community table, @Bruce !