p. bhogale

Companion to "Interrogating Your Twin: Causal Reasoning in Manufacturing Systems" (Fifth Elephant 2026, Pune Edition). Earlier posts in this series:

1. Climbing Pearl's Ladder of Causation --- the conceptual foundations
2. A causal workflow with coupon marketing data --- the five-step pipeline applied to marketing
3. This post --- real factory data, reverse causation, and the …

This post is code for the demo associated with a talk on causal inference presented at Fifth Elephant Winter 2025. The video of the talk will be available soon, and I'll link to it here.

We walk through a complete causal inference workflow using real marketing data from the AmExpert …

Disclaimer: statistics is hard - the chief skill seems to be the ability to avoid deluding oneself and others. This is something that is best and quickest learned via an apprenticeship in a group of careful thinkers trying to get things right. Tutorials like these can be misleading, in that they …

Reading material

1. Please read the previous blogpost for basics of conformal inference and more reading material.
2. This post makes use of the conformalForecast package introduced in the paper published just 10 days ago by Wang and Hyndman.
3. For a more general overview of time series forecasting, please see this excellent …

Reading material:

1. The tidy modeling book
2. The tidymodels blog on conformal regression
3. The notes of Angelopoulos
4. The notes of Tibshirani
5. The book of Christoph Molnar
6. The book of Valeriy Manokhin
7. The package of Sussman et al.

Getting some data

We will look at Indian trade data hosted on Kaggle for …

These notes are meant to implement little examples from Francois Fleuret's Little Book of Deep Learning (pdf link) in r-torch. I'm writing these as a fun way to dive into torch in R while surveying DL quickly.

You'll need to have the book with you to understand these notes, but …

There are many articles on this subject, and many of them are excellent. For some examples see here, here, here and here. All such articles are opinionated takes, and articles like these did help when I was starting out in data science some years ago. Contrary to my expectations, the …

post inspired by the OptiPy meetup. link to meetup

Linear programming

Optimisation problems where the constraints and cost function are linear, and the decision variables are continuous, are the simple, canonical domain of linear programming. However, such problems can be solved in polynomial time, which means that to tackle a …

Markets and financial time series data.

Basic ideas about markets

Inevitably, finance and markets will be a recurring subject of discussion on this blog. They are ubiquitous, useful, fascinatingly complex at various levels and they generate a lot of readily accessible data.

What is a market ? A market is a …

A first foray into probabilistic programming with Greta

Models and modelling

Much of science - physical and social - is devoted to positing mechanisms that explain how the data we observe are generated. In a classic example, after Tycho Brahe made detailed observations of planetary motion (here is data on mars), Johannes …

These are notes I took while watching the lectures from Andrew Ng's ML course. There is no code, just some math and my take aways from the course. It should still serve as a useful first document to skim for someone just starting out with machine learning.

What can ML …

And so there she is, our baby daughter. Even after a fairly uneventful pregnancy, the birth of a baby must be among the most epic things men can witness and women can experience. The pain that women go through alone makes the experience singular, even without a whole new person …

1.

"I've checked", the Principal said one morning at assembly "The hill behind the school does not seem to have a name. In the memory of those who died in Kashmir, we will call it Tiger Hill. So that we never forget."

It was the monsoon of 1999. I stood …