diff options
Diffstat (limited to 'posts')
| -rw-r--r-- | posts/2018-12-02-lime-shapley.md | 28 |
1 files changed, 14 insertions, 14 deletions
diff --git a/posts/2018-12-02-lime-shapley.md b/posts/2018-12-02-lime-shapley.md index 88b1b20..add6881 100644 --- a/posts/2018-12-02-lime-shapley.md +++ b/posts/2018-12-02-lime-shapley.md @@ -8,7 +8,7 @@ comments: true In this post I explain LIME (Ribeiro et. al. 2016), the Shapley values (Shapley, 1953) and the SHAP values (Lundberg-Lee, 2017). -Shapley values {#Shapley values} +Shapley values -------------- A coalitional game $(v, N)$ of $n$ players involves @@ -37,14 +37,14 @@ $i$. The Shapley values satisfy some nice properties which are readily verified, including: -- []{#Efficiency}**Efficiency**. +- **Efficiency**. $\sum_i \phi_i(v) = v(N) - v(\emptyset)$. -- []{#Symmetry}**Symmetry**. If for some $i, j \in N$, for all +- **Symmetry**. If for some $i, j \in N$, for all $S \subset N$, we have $v(S + i) = v(S + j)$, then $\phi_i(v) = \phi_j(v)$. -- []{#Null player}**Null player**. If for some $i \in N$, for all +- **Null player**. If for some $i \in N$, for all $S \subset N$, we have $v(S + i) = v(S)$, then $\phi_i(v) = 0$. -- []{#Linearity}**Linearity**. $\phi_i$ is linear in games. That is +- **Linearity**. $\phi_i$ is linear in games. That is $\phi_i(v) + \phi_i(w) = \phi_i(v + w)$, where $v + w$ is defined by $(v + w)(S) := v(S) + w(S)$. @@ -61,7 +61,7 @@ Conversely, according to Strumbelj-Kononenko (2010), it was shown in Shapley\'s original paper (Shapley, 1953) that these four properties together with $v(\emptyset) = 0$ defines the Shapley values. -LIME {#LIME} +LIME ---- LIME (Ribeiro et. al. 2016) is a model that offers a way to explain @@ -100,7 +100,7 @@ of perturbed samples $x_i = h_x(S_i)$ in the $X$ space, and then fits a linear model $g$ using $1_{S_i}$ as the input samples and $f(h_x(S_i))$ as the output samples: -[]{#Problem}**Problem**(LIME). Find $w = (w_1, w_2, ..., w_n)$ that +**Problem**(LIME). Find $w = (w_1, w_2, ..., w_n)$ that minimises $$\sum_i (w \cdot 1_{S_i} - f(h_x(S_i)))^2 \pi_x(h_x(S_i))$$ @@ -142,11 +142,11 @@ $\sum_{i = 1}^n w_i = f(x) - f(h_x(\emptyset)) = v(N) - v(\emptyset)$. Then the problem becomes a weighted linear regression: -[]{#Problem}**Problem**. minimises +**Problem**. minimises $\sum_{S \subset N} (w(S) - v(S))^2 q(s)$ over $w$ subject to $w(N) = v(N) - v(\emptyset)$. -[]{#Claim}**Claim** (Charnes et. al. 1988). The solution to this problem +**Claim** (Charnes et. al. 1988). The solution to this problem is $$w_i = {1 \over n} (v(N) - v(\emptyset)) + \left(\sum_{s = 1}^{n - 1} {n - 2 \choose s - 1} q(s)\right)^{-1} \sum_{S \subset N: i \in S} \left({n - s \over n} q(s) v(S) - {s - 1 \over n} q(s - 1) v(S - i)\right). \qquad (-1)$$ @@ -157,7 +157,7 @@ $$q(s) = c {n - 2 \choose s - 1}^{-1}$$ for any constant $c$, then $w_i = \phi_i(v)$ are the Shapley values. -[]{#Remark}**Remark**. Don\'t worry about this specific choice of $q(s)$ +**Remark**. Don\'t worry about this specific choice of $q(s)$ when $s = 0$ or $n$, because $q(0)$ and $q(n)$ do not appear on the right hand side of (-1). Therefore they can be defined to be of any value. A common convention of the binomial coefficients is to set @@ -167,7 +167,7 @@ $q(0) = q(n) = \infty$. In Lundberg-Lee (2017), $c$ is chosen to be $1 / n$, see Theorem 2 there. -[]{#Proof}**Proof**. The Lagrangian is +**Proof**. The Lagrangian is $$L(w, \lambda) = \sum_{S \subset N} (v(S) - w(S))^2 q(s) - \lambda(w(N) - v(N) + v(\emptyset)).$$ @@ -219,7 +219,7 @@ $$\sum_{S \subset N} s q(s) v(S) = \sum_{S: i \in S} (s q(s) v(S) + (s - 1) q(s Plugging this back into (6) we get the desired result. $\square$ -SHAP {#SHAP} +SHAP ---- The SHAP paper (Lundberg-Lee 2017) is not clear in its definition of the @@ -266,12 +266,12 @@ x_i, & \text{if }i \in S; \\ \mathbb E_{\mu_i} z_i, & \text{otherwise.} \end{cases}$$ -How about DeepLIFT? {#How about DeepLIFT?} +How about DeepLIFT? ------------------- TODO -References {#References} +References ---------- - Charnes, A., B. Golany, M. Keane, and J. Rousseau. "Extremal |