diff options
Diffstat (limited to 'posts/2018-12-02-lime-shapley.org')
| -rw-r--r-- | posts/2018-12-02-lime-shapley.org | 6 |
1 files changed, 6 insertions, 0 deletions
diff --git a/posts/2018-12-02-lime-shapley.org b/posts/2018-12-02-lime-shapley.org index 05ef4ee..0eaacf2 100644 --- a/posts/2018-12-02-lime-shapley.org +++ b/posts/2018-12-02-lime-shapley.org @@ -17,6 +17,7 @@ under CC BY-SA and GNU FDL./ ** Shapley values :PROPERTIES: :CUSTOM_ID: shapley-values + :ID: 1f4de00d-669e-421a-bd48-f4f1da8400f3 :END: A coalitional game $(v, N)$ of $n$ players involves @@ -69,6 +70,7 @@ together with $v(\emptyset) = 0$ defines the Shapley values. ** LIME :PROPERTIES: :CUSTOM_ID: lime + :ID: b34aabbc-2f50-4969-bf7b-87678fd577e6 :END: LIME (Ribeiro et. al. 2016) is a model that offers a way to explain feature contributions of supervised learning models locally. @@ -139,6 +141,7 @@ of dimension $n$. ** Shapley values and LIME :PROPERTIES: :CUSTOM_ID: shapley-values-and-lime + :ID: bb960a6a-be79-44ff-968f-000990673738 :END: The connection between the Shapley values and LIME is noted in Lundberg-Lee (2017), but the underlying connection goes back to 1988 @@ -243,6 +246,7 @@ Plugging this back into (6) we get the desired result. $\square$ ** SHAP :PROPERTIES: :CUSTOM_ID: shap + :ID: 11af7fd2-a4f9-42da-a275-5d15cde0ca48 :END: The paper that coined the term "SHAP values" (Lundberg-Lee 2017) is not clear in its definition of the "SHAP values" and its relation to LIME, @@ -304,6 +308,7 @@ general, how do we evaluate models of interpretation? ** Evaluating SHAP :PROPERTIES: :CUSTOM_ID: evaluating-shap + :ID: a7249300-814f-499e-b808-c66a17a4d32e :END: The quest of the SHAP paper can be decoupled into two independent components: showing the niceties of Shapley values and choosing the @@ -332,6 +337,7 @@ justifications of such choices. ** References :PROPERTIES: :CUSTOM_ID: references + :ID: 7ba9546e-2485-43e4-be48-16762fc447b2 :END: - Charnes, A., B. Golany, M. Keane, and J. Rousseau. "Extremal Principle |