Update Shapley.md

This commit is contained in:
Nuño Sempere 2019-09-12 12:31:40 +02:00 committed by GitHub
parent 45630e4c42
commit c6dee0f890
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

View File

@ -66,7 +66,7 @@ Value(AS) = 1/3 * [(Value of adding him to {} / 1) + (Value of adding him to {DB
Value of adding him to {} = 0
Number of coalitions excluding him of size 0 = 1
Value of adding him to {DB} = 0
Value of adding him to {MS} = 0.4 ## The mad scientist tries to talk to him, but he's not very convincing
Value of adding him to {MS} = 0.4 // The mad scientist tries to talk to him, but he's not very convincing
Number of coalitions excluding him of size 1 = 2
Value of adding him to {MS, DB} = 1
Number of coalitions excluding him of size 2 = 1
@ -86,7 +86,7 @@ Number of coalitions excluding him of size 0 = 1
Value of adding him to {AS} = 0
Value of adding him to {MS} = 0
Number of coalitions excluding him of size 1 = 2
Value of adding him to {MS, AS} = 1-0.4 = 0.6 ## These are MARGINAL values.
Value of adding him to {MS, AS} = 1-0.4 = 0.6 // These are MARGINAL values.
Number of coalitions excluding him of size 2 = 1
Value(AS) = 1/3 * (0/1 + 0/2 + 0/2 + 0.6/1) = 0.2
@ -243,9 +243,11 @@ end <- Sys.time()
end-start
sum(2/(c(1:1000))^2)```
```
## Sources
https://en.wikipedia.org/wiki/Shapley_value
https://christophm.github.io/interpretable-ml-book/shapley.html
https://towardsdatascience.com/understanding-how-ime-shapley-values-explains-predictions-d75c0fceca5a
https://towardsdatascience.com/one-feature-attribution-method-to-supposedly-rule-them-all-shapley-values-f3e04534983d
https://www.eeg.tuwien.ac.at/conference/iaee2017/files/presentation/Pr_662_Jamasb_Tooraj.pdf (This last one is an interesting use fo Shapley values for international coordination)