Michael Knaus：因果机器学习课件

[chengjun]

Michael Knaus：因果机器学习

来自: 苏格拉底大王(I am serious with Socrates.) 组长 2023-03-16 16:50:15 江苏

德国图宾根大学助理教授Michael Knaus给喜欢因果机器学习的经济学者分享了一份学习材料。

这里整合了Knaus教授给他的硕士和博士生教授的几门“因果机器学习”课程的材料，重点关注的是政策影响/项目评估。

它里面包括10份Slides，对应着因果机器学习的十个主题，22份R notebooks。

Teaching material for Causal ML

This repository consolidates the teaching material of several "Causal Machine Learning" courses I taught on the master and PhD level with a focus on impact/policy/program evaluation.

Comments

Like the whole literature the content is a moving target. Please let me know if you spot any errors, disagreements, but also if you found the material useful. To this end, open an issue or write me a mail

The slides include links to a variety of compiled html R notebooks. Their Rmd files are provided in this repository if you are iterested in running and extending them yourself. A full list of available notebooks is provided on my homepage.

Slides

1. Welcome
2. Stats/’metrics recap
3. Supervised ML: predicting outcomes
4. Causal Inference basis
5. Estimating constant effects: Double Selection to Double ML
6. Average treatment effect estimation: AIPW-Double ML
7. Double ML - the general recipe
8. Predicting effects
9. Heterogeneous effects with inference
10. Policy learning

在他个人主页里还有更多因果机器学习相关素材值得仔细研读。

Simulation notebooks

These R Notebooks illustrate concepts and methods with simulated data. You can extract the code by clicking on the top right button and download the Rmd file to run/modify the code.

• Basics: Convergence rates
• Supervised ML: Overfitting of OLS and value of training vs. test sample
• Supervised ML: Lasso saves the job of OLS
• Supervised ML: Tree-based methods
• Causal ML: Why naive model selection fails
• Causal ML: Double Selection
• Causal ML: Partially linear Double ML
• Causal ML: AIPW Double ML (ATE)
• Causal ML: Group Average Treatment Effects
• Causal ML: Causal Tree and Causal Forest
• Causal ML: Meta-learner
• Causal ML: Offline policy learning
• Causal ML: Multi-armed bandit

Application notebooks

These R notebooks use the pension dataset of Chernozhukov and Hansen (2004) that is part of the hdm package. Focus is on hand-coding the methods to see how they work.

• Basics: OLS and Frisch-Waugh
• Supervised ML: Lasso
• Supervised ML: Tree-based methods
• Causal ML: Double Selection and Partially Linear Double ML
• Causal ML: Double ML for average treatment effects
• Causal ML: Double ML as generic recipe
• Causal ML: Double ML for group average treatment effects
• Causal ML: Predicting effects
• Causal ML: Offline policy learning

Identification notebooks

Showcase the use of DAGs and SWIGs.

• Causal Inference: DAG and SWIG for RCTs
• Causal Inference: DAG and SWIG for measured confounding

Knaus教授讲解非常生动、形象、有趣，从日常生活中简单的事情逐步深入到因果关系以及因果机器学习的核心内容。

下面就是因果推断基础的部分Slides，是不是很形象易懂？

下载网址：https://github.com/MCKnaus/causalML-teaching

Knaus教授主页：https://mcknaus.github.io/menu/teaching.html