"Causal Inference & Discovery in Python" is out now!
Foreword by
@AjitJaokar
(
@UniofOxford
)
We did our best to avoid mistakes in the book but some happened
Here's an errata file with known mistakes:
If you spot more, let me know
Some people say that causal modeling requires too many assumptions.
And it's better to use associative modeling, because it requires less .
#causality
#machinelearning
#causaltwitter
1/n
❓ What are the sources of causal knowledge?
- Randomized experiments
- Personal expertise / domain knowledge
- Causal discovery / structure learning
Am I missing something important?
#causality
#causaltwitter
I adapted the code of chapter 3 of
@CausalHuber
's book to Jupyter notebook to make it more accessible for the Python community.
I introduced some minor improvements and added some visuals.
You can check it here:
🟠 Here's a brand new causal blog post 👇🏼
Guido Imbens & Susan Athey called this method "arguably the most important innovation in the policy evaluation literature in the last 15 years".
#causaltwitter
#causality
#python
#machinelearning
⭕ 𝗖𝗮𝘂𝘀𝗮𝗹𝗣𝘆 ()
A new kid on the block. CausalPy is based on
@pymc_labs
PyMC. Developed by PyMC’s prolific
@inferencelab
with contributions from
@juanitorduz
. CausalPy implements a bunch of methods to work with quasi-experimental data.
(6/6)
Preparing for causal 2024?
Here's my take on the Python's causal ecosystem from Q1 2023.
Which packages (if any) should we add to the mix for 2024?
1/n
🥷🏼 Here's a casual concept that even seasoned scientists & senior data scientists struggle with
...and 2 fresh resources that will help you master it 👇🏼👇🏼
🧵
💡 The code snippet presents an SCM model that we use in chapter 6 of my upcoming causal book
The room was almost completely full until the last minute.
Our participants shared their thoughts and insights, sparking more engaging discussions.
A summary tomorrow.
At our Causal Parrots Workshop
@RealAAAI
#aaai2024
@devendratweetin
@matej_zecevic
This is also one of the reasons why non-causal approaches to root cause analysis fail.
High variation is a (only) a (fallible) proxy of a factor's contribution importance.
First, many analyses (regression, PCA) assume high variance representation components are most important. But equally-important features may not carry equal variance; e.g. if a model computes easy (linear) and hard (nonlinear) tasks, the easy one dominates the representations! 3/
Ready to jump-start your causal inference journey?
@AleksanderMolak
is here to help, with a clear, Python-based guide that explains how to estimate the causal effect of one variable on another variable from observational data.
Yesterday, in Edinburgh, we met with
@stephensenn
for a conversation about randomization, causal inference, and challenges and opportunities in clinical trials and drug development.
#causality
#causaltwitter
#rct
1/n
In their 2020 paper (),
@analisereal
and C. Hazlett from UCLA proposed a flexible method** that does not require assumptions on the functional form of the treatment assignment nor on the distribution of the unobserved confounders
🧵 (9/n)
@MattWalshBlog
@MattWalshBlog
controlling for "all relevant variables" is a bad startegy in most real-life analyses as it likely results in controlling for colliders, mediators or their descendants. Such controls will lead to arbitrary biases in the estimates, including sign flipping
✌🏼2️⃣ There are two causal concepts that will bend your statistical mind if you haven't heard about them before.
(thrd🧵)
#causality
#causaltwitter
#Python
#machinelearning
💡 The code snippet below with comes from chapter 5 of my upcoming causal book: