Slide 1

Slide 1 text

2022/2/17 Developers Summit 2022 @dropout009

Slide 2

Slide 2 text

TVISION INSIGHTS Twitter: @dropout009 Speaker Deck: dropout009 Blog: https://dropout009.hatenablog.com/

Slide 3

Slide 3 text

Deep Learning GBDT Random Forest

Slide 4

Slide 4 text

No content

Slide 5

Slide 5 text

画像認識、⾳声認識、機械翻訳、… Deep Learning, Gradient Boosting Decision Tree, Random Forest, ... pytorch, LightGBM, scikit-learn, ...

Slide 6

Slide 6 text

LM GBDT RF DL GLM

Slide 7

Slide 7 text

No content

Slide 8

Slide 8 text

LM GBDT RF DL GLM +

Slide 9

Slide 9 text

No content

Slide 10

Slide 10 text

4 PFI PD + ICE SHAP

Slide 11

Slide 11 text

1. PFI 2. PD 3. ICE 4. SHAP 住宅価格 = 𝛽! + 𝛽" 部屋の数 + 𝛽# 駅からの距離 + ノイズ

Slide 12

Slide 12 text

Partial Dependence PD 𝑌 = 𝑋! − 5𝑋" + 10𝑋"𝑋# + 𝜖 𝑋!, 𝑋" ∼ Uniform −1, 1 𝑋# ∼ Bernoulli 0.5 𝜖 ∼ 𝒩(0, 0.01)

Slide 13

Slide 13 text

Python Partial Dependence @dataclass class PartialDependence: """Partial Dependence (PD)""" estimator: Any X: np.ndarray var_names: list[str] def _counterfactual_prediction( self, idx_to_replace: int, value_to_replace: float ) -> np.ndarray: """ある特徴量の値を置き換えたときの予測値を求める""" X_replaced = self.X.copy() X_replaced[:, idx_to_replace] = value_to_replace y_pred = self.estimator.predict(X_replaced) return y_pred ……… $ PD> 𝑥> = 1 𝑁 * ?@" A + 𝑓(𝑥> , 𝒙?,∖> )

Slide 14

Slide 14 text

No content