Slide 26
Slide 26 text
LIME
•lime()にデータとモデルを食わせる
•explain()で指定したデータの局所解釈を作成
• glmnet(L1+L2なロジスティック回帰)で説明
case label feature feature_weight feature_desc data
1 116 virginica Sepal.Length 0.0095341177 5.8 < Sepal.Length <= 6.4 6.4, 3.2, 5.3, 2.3, 3.0
2 116 virginica Petal.Length 0.4583012101 5.10 < Petal.Length 6.4, 3.2, 5.3, 2.3, 3.0
3 36 setosa Species 0.0008123117 Species = setosa 5.0, 3.2, 1.2, 0.2, 1.0
4 36 setosa Petal.Length 0.4377915136 Petal.Length <= 1.60 5.0, 3.2, 1.2, 0.2, 1.0
5 4 setosa Sepal.Width -0.0010979380 3.0 < Sepal.Width <= 3.3 4.6, 3.1, 1.5, 0.2, 1.0
6 4 setosa Petal.Length 0.4591607286 Petal.Length <= 1.60 4.6, 3.1, 1.5, 0.2, 1.0
7 111 virginica Species 0.0042950996 Species = virginica 6.5, 3.2, 5.1, 2.0, 3.0
8 111 virginica Petal.Width 0.4008624393 1.8 < Petal.Width 6.5, 3.2, 5.1, 2.0, 3.0
9 37 setosa Sepal.Length 0.0097798516 5.1 < Sepal.Length <= 5.8 5.5, 3.5, 1.3, 0.2, 1.0
10 37 setosa Petal.Length 0.4625155191 Petal.Length <= 1.60 5.5, 3.5, 1.3, 0.2, 1.0