Try   HackMD

5.3 Accumulated Local Effects (ALE) Plot

M-Plots

  • 條件機率
  • 參雜其他相關變數的效果

ALE Plots

  • 依照觀察變數的範圍，切成N段(Intervals)
  • 將每個instances的變數值帶入所在區間的最大值和最小值，求其差
  • 除以區間內的樣本數 –> 中心化 –> 相加

ALE plots for 變數間的交互作用項

  • Second-order effect : 只考慮交互作用項帶來的額外效果

ALE plots for 類別變數

  • 計算每個class之間的similarity –> 得到Distance Matrix –> 再利用MDS(Multi-Dimensional Scaling)將所有資料點降至一維
  • 如何計算Similarity:

Differences

  • Partial Dependence Plots: “Let me show you what the model predicts on average when each data instance has the value v for that feature. I ignore whether the value v makes sense for all data instances.”
  • M-Plots: “Let me show you what the model predicts on average for data instances that have values close to v for that feature. The effect could be due to that feature, but also due to correlated features.”
  • ALE plots: “Let me show you how the model predictions change in a small”window" of the feature around v for data instances in that window."

Advantages

  • 可以處理有關聯性的變數
  • 計算速度較PDP快
  • 解讀較易懂，由於ALE plots集中在0,條件機率下因子的變化較為明顯

Disadvantages

  • ALE plots can become a bit shaky (many small ups and downs) with a high number of intervals.
  • 無法看到每一個因子異質性的影響效果
  • Second-order ALE 在不同feature 隔窗的穩定性差異大，因為每一的隔窗的資料是不同的
  • Second-order effect plots can be a bit annoying to interpret
  • implementation of ALE plots is much more complex也較PDP不直覺

Use ALE instead of PDP.

參考資料

Python Code

tags: 重點摘要
Last changed by 
 
Explainable AI
0
373

Read more

NLP xai

https://arxiv.org/pdf/2010.00711.pdf 總結目前NLP相關paper使用的解釋方法有哪些 Explanation derivation -- 數學上的解釋 Feature importances Explanation presentation -- 如何呈現這些Explanation Visualization

Dec 24, 2020
Questioning the AI: Informing Design Practices for Explainable AI User Experiences

Dec 9, 2020
讀書會時程規劃

閱讀書籍 - Interpretable Machine Learning A Guide for Making Black Box Models Explainable. 兩篇論文： https://arxiv.org/pdf/2001.02478.pdf https://arxiv.org/pdf/1909.03012.pdf 書籍網址 https://christophm.github.io/interpretable-ml-book/ 倚任副理建議直接先閱讀第五章和第六章 讀書計畫初稿如下表，從下週二開始分享，之後視情況可機動調整

Nov 26, 2020
6.4 Influential Instances

機器學習的模型是訓練數據的產出，刪除任一訓練數據會影響訓練結果。若刪除每一訓練數據對模型產生巨大影響，則稱這個點唯有影響的點(instance)。對有影響的點分析可以幫助我們檢視模型。 Deletion Diagnostics : delete the instance from the training data, retrain the model on the reduced training dataset and observe the difference in the model parameters or predictions Influenced functions : upweight a data instance by approximating the parameter changes based on the gradients of the model parameters. 6.4.1 Deletion Diagnostics DFBETA : 衡量移除某個instance對模型參數的影響。 $DFBETA_i = β-β^{-i}$ 適用於有參數的模型，如 logistic regression or neural networks.

Nov 6, 2020
Read more from Explainable AI
Published on HackMD