al (2016) , the Shapley Value described by Strumbelj et. iml is an R package that interprets the behavior and explains predictions of machine learning models. The Shapley value fairly distributes the difference of the instance's prediction and the datasets average prediction among the features. Another approach is called breakDown, which is implemented in the breakDown R package 40. We can also fit a model locally to understand an individual prediction better. Shapley computes feature contributions for single predictions with the Shapley value, an approach from cooperative game theory. The iml package provides tools for analysing any black box machine learning model:Surrogate trees: Can we approximate the underlying black box model with a short decision tree?We can reuse the object to explain other data points:Explanations for single predictions: How did the feature values of a single data point affect its prediction? (LIME and Shapley value)Besides knowing which features were important, we are interested in how the features influence the predicted outcome. Description. Qiita can be used more conveniently after logging in.Help us understand the problem. plot.Shapley: Plot Shapley in iml: Interpretable Machine Learning rdrr.io Find an R package R language docs Run R in your browser R Notebooks Vignettes. #' mod <- Predictor$new(rf, data = X, type = "prob", class = "setosa")#' `Shapley` computes feature contributions for single predictions with the#' X <- Boston[-which(names(Boston) == "medv")]#' The number of times coalitions/marginals are#' Strumbelj, E., Kononenko, I. Parallel computation of interpretation methods Shapley values are implemented in both the iml and fastshap packages for R. SHAP, an alternative estimation method for Shapley values, is presented in the next chapter. Explaining prediction models and iml: Interpretable Machine Learning. Name : Description : Partial: Effect of one or two feature(s) on the model predictions (deprecated) FeatureImp: Feature importance: InterpretationMethod Shapley computes feature contributions for single predictions with the Shapley value, an approach from cooperative game theory. It provides both global and local model-agnostic interpretation methods. Shapley computes feature contributions for single predictions with the Shapley value, an approach from cooperative game theory. The Shapley value tells us how to fairly distribute the payout among the feature values. Introduction to iml: Interpretable Machine Learning in R Introduction to iml: Interpretable Machine Learning in R Hi there! Explaining prediction models and#' # Then we explain the first instance of the dataset with the Shapley method:#' Single row with the instance to be explained.#' Systems, 41(3), 647-665. https://doi.org/10.1007/s10115-013-0679-x#' Should the feature values be sorted by Shapley value? Qiita can be used more conveniently after logging in.#7 ptratio 1.8601196 2.006714 2.251311 4.125215X軸のマークは'lstat'の分布を表しており、どの領域が関係性が高いかを示しています。#4 0.03237 0 2.18 0 0.458 6.998 45.8 6.0622 3 222 18.7 394.63 2.94#2 rm 10.9331383 11.489301 19.441368 23.618625#6 rm -0.450538504 15.837141138 rm=6.575#ptratio -0.5244767 15.300 -8.024493 15.3 ptratio ptratio=15.3# feature phi phi.var feature.value#rm 4.4836483 6.575 29.479987 6.575 rm rm=6.575#12 chas 1.0462975 1.126355 1.171892 2.315455今後も説明可能なAIの需要は高まっていき、次々と新しい理論・実装が現れてくることが予想されます。ランダムフォレストモデルの作成は目的変数、使用データセット、木の深さを指定するだけで簡単にできます。# self$predictor$data$match_cols(data.frame(newdata)) で: #6 0.02985 0 2.18 0 0.458 6.430 58.7 6.0622 3 222 18.7 394.12 5.21How to use in R model-agnostic data explanation with DALEX & iml#8 age 1.5103398 1.827092 2.179101 3.755965#1 0.00632 18 2.31 0 0.538 6.575 65.2 4.0900 1 296 15.3 396.90 4.98#5 nox -0.151510825 1.830613219 nox=0.538#4 nox 2.8722394 3.003836 3.498187 6.175004H-statisticの詳しい説明はここでは割愛しますが、H-statisticは[0,1]の無次元量であるため、異なる特徴量・モデル間の比較に使うことができます。ゲーム理論を使って各変数の貢献度を計算しよう、という少し今までとはテイストの違う考え方がShapley Valueです。#6 indus 1.8486110 2.099748 2.274380 4.316464この例ではランダムフォレストを決定木に変換していますが、あらゆるブラックボックスモデルに適用可能です。#1 crim -0.317960865 2.046510169 crim=0.00632近年のAI・機械学習ブームの一方で、一部の予測モデルのブラックボックス的な性質ゆえに、AIへの過剰な期待・無責任なAI利用への警鐘を鳴らす声も頻繁に聞かれるようになりました。Introduction to iml: Interpretable Machine Learning in RSign up for free and join this conversation.ALEと似た概念にPDP (Partial Dependence Plot)がありますが、PDPでは「あるデータに対して、注目する特徴量（変数）以外はすべて同じ値をもつデータがたくさんあったとき、対象の特徴量（変数）の値の増減によって予測値がどう変化するか？」ということを調べます。