Questing for “the scikit learn of causal inference”

Questing for “the scikit learn of causal inference”

it has a simple modular workflow

it supports a wide variety of models under a common interface.

it solves a task people frequently have, ie E[y

X]. There are multiple examples of this in Casual Inference

Obtaining the causal graph

There are many ways to do this

• Discovery algorithms
• Domain knowledge
• Experiments
• Econometric analysis
• Regression

Then represent it as networkx, maybe also summarize with a single number related to the arrow strength or elasticity or coefficient or interventional_samples

An example workflow and library: DoWhy

DoWhy close reading - “An opinionated introduction to DoWhy”

https://www.youtube.com/watch?v=icpHrbDlGaw

I. Model a causal problem

Currently, dowhy doesn’t really support discovery. Arrow strength is useful. Maybe include an example from something else here

II. Identify a target estimand under the model

III. Estimate causal effect based on the identified estimand

IV. Refute the obtained estimate

Pick an example including both a root cause analysis and an intervention analysis from the R data sets.

• House pricing? RCA: Why so expensive; Intervention: what if I made some renovation;

Assembling the causal graph

• The DAG (NetworkX)
• Root distributions and interior functional forms
• How do we deal with categorical variables? (Look at the docs for this)
• How to fit the model
• Root cause analysis howto
• Effect estimation