fixes #328

Author: christophM

README.md

@@ -110,11 +110,12 @@ All notable changes to the book will be documented here.
 - Renamed Feature Importance chapter to "Permutation Feature Importance"
 - Added chapter about functional decomposition
 - Rearranged interpretation methods by local, global and deep learning (before: model-agnostic, example-based, deep learning)
-- Errata:
+- Math Errata:
 	- Chapter 4.3 GLM, GAM and more: Logistic regression uses logit, not logistic function as link function.
 	- Chapter Linear models: Formula for adjusted R-squared was corrected (twice)
         - Chapter Decision Rules: Newly introduced mix up between Healthy and Cancer in OneR chapter was fixed.
         - Chapter RuleFit: The importance of the linear term in the total importance formulate was indexed with an $l$ instead of $j$.
+    - Chapter Influential Instances: removed $(1-\epsilon)$ from model parameter update.
 - Updated images
 
 ### v1.1 (2019-03-23) 

manuscript/influential.qmd

@@ -334,7 +334,7 @@ The following section explains the intuition and math behind influence functions
 
 The key idea behind influence functions is to upweight the loss of a training instance by an infinitesimally small step $\epsilon$, which results in new model parameters:
 
-$$\hat{\theta}_{\epsilon, \mathbf{z}} = \arg\min_{\theta \in \Theta} \left( (1 - \epsilon)\frac{1}{n}\sum_{i=1}^n L(\mathbf{z}^{(i)}, \theta) + \epsilon L(\mathbf{z}, \theta) \right)$$
+$$\hat{\theta}_{\epsilon, \mathbf{z}} = \arg\min_{\theta \in \Theta} \left(\frac{1}{n}\sum_{i=1}^n L(\mathbf{z}^{(i)}, \theta) + \epsilon L(\mathbf{z}, \theta) \right)$$
 
 where $\theta$ is the model parameter vector and $\hat{\theta}_{\epsilon, \mathbf{z}}$ is the parameter vector after upweighting $\mathbf{z}$ by a very small number $\epsilon$.
 $L$ is the loss function with which the model was trained, $\mathbf{z}^{(i)}$ is the training data, and $\mathbf{z}$ is the training instance which we want to upweight to simulate its removal.